☆ Are Video Models Ready as Zero-Shot Reasoners? An Empirical Study with the MME-CoF Benchmark

Ziyu Guo, Xinyan Chen, Renrui Zhang, Ruichuan An, Yu Qi, Dongzhi Jiang, Xiangtai Li, Manyuan Zhang, Hongsheng Li, Pheng-Ann Heng

Recent video generation models can produce high-fidelity, temporally coherent videos, indicating that they may encode substantial world knowledge. Beyond realistic synthesis, they also exhibit emerging behaviors indicative of visual perception, modeling, and manipulation. Yet, an important question still remains: Are video models ready to serve as zero-shot reasoners in challenging visual reasoning scenarios? In this work, we conduct an empirical study to comprehensively investigate this question, focusing on the leading and popular Veo-3. We evaluate its reasoning behavior across 12 dimensions, including spatial, geometric, physical, temporal, and embodied logic, systematically characterizing both its strengths and failure modes. To standardize this study, we curate the evaluation data into MME-CoF, a compact benchmark that enables in-depth and thorough assessment of Chain-of-Frame (CoF) reasoning. Our findings reveal that while current video models demonstrate promising reasoning patterns on short-horizon spatial coherence, fine-grained grounding, and locally consistent dynamics, they remain limited in long-horizon causal reasoning, strict geometric constraints, and abstract logic. Overall, they are not yet reliable as standalone zero-shot reasoners, but exhibit encouraging signs as complementary visual engines alongside dedicated reasoning models. Project page: https://video-cof.github.io

comment: Project Page: https://video-cof.github.io

☆ Gistify! Codebase-Level Understanding via Runtime Execution

Hyunji Lee, Minseon Kim, Chinmay Singh, Matheus Pereira, Atharv Sonwane, Isadora White, Elias Stengel-Eskin, Mohit Bansal, Zhengyan Shi, Alessandro Sordoni, Marc-Alexandre Côté, Xingdi Yuan, Lucas Caccia

As coding agents are increasingly deployed in large codebases, the need to automatically design challenging, codebase-level evaluation is central. We propose Gistify, a task where a coding LLM must create a single, minimal, self-contained file that can reproduce a specific functionality of a codebase. The coding LLM is given full access to a codebase along with a specific entrypoint (e.g., a python command), and the generated file must replicate the output of the same command ran under the full codebase, while containing only the essential components necessary to execute the provided command. Success on Gistify requires both structural understanding of the codebase, accurate modeling of its execution flow as well as the ability to produce potentially large code patches. Our findings show that current state-of-the-art models struggle to reliably solve Gistify tasks, especially ones with long executions traces.

☆ Defeating the Training-Inference Mismatch via FP16

Penghui Qi, Zichen Liu, Xiangxin Zhou, Tianyu Pang, Chao Du, Wee Sun Lee, Min Lin

Reinforcement learning (RL) fine-tuning of large language models (LLMs) often suffers from instability due to the numerical mismatch between the training and inference policies. While prior work has attempted to mitigate this issue through algorithmic corrections or engineering alignments, we show that its root cause lies in the floating point precision itself. The widely adopted BF16, despite its large dynamic range, introduces large rounding errors that breaks the consistency between training and inference. In this work, we demonstrate that simply reverting to \textbf{FP16} effectively eliminates this mismatch. The change is simple, fully supported by modern frameworks with only a few lines of code change, and requires no modification to the model architecture or learning algorithm. Our results suggest that using FP16 uniformly yields more stable optimization, faster convergence, and stronger performance across diverse tasks, algorithms and frameworks. We hope these findings motivate a broader reconsideration of precision trade-offs in RL fine-tuning.

☆ Remote Labor Index: Measuring AI Automation of Remote Work

Mantas Mazeika, Alice Gatti, Cristina Menghini, Udari Madhushani Sehwag, Shivam Singhal, Yury Orlovskiy, Steven Basart, Manasi Sharma, Denis Peskoff, Elaine Lau, Jaehyuk Lim, Lachlan Carroll, Alice Blair, Vinaya Sivakumar, Sumana Basu, Brad Kenstler, Yuntao Ma, Julian Michael, Xiaoke Li, Oliver Ingebretsen, Aditya Mehta, Jean Mottola, John Teichmann, Kevin Yu, Zaina Shaik, Adam Khoja, Richard Ren, Jason Hausenloy, Long Phan, Ye Htet, Ankit Aich, Tahseen Rabbani, Vivswan Shah, Andriy Novykov, Felix Binder, Kirill Chugunov, Luis Ramirez, Matias Geralnik, Hernán Mesura, Dean Lee, Ed-Yeremai Hernandez Cardona, Annette Diamond, Summer Yue, Alexandr Wang, Bing Liu, Ernesto Hernandez, Dan Hendrycks

AIs have made rapid progress on research-oriented benchmarks of knowledge and reasoning, but it remains unclear how these gains translate into economic value and automation. To measure this, we introduce the Remote Labor Index (RLI), a broadly multi-sector benchmark comprising real-world, economically valuable projects designed to evaluate end-to-end agent performance in practical settings. AI agents perform near the floor on RLI, with the highest-performing agent achieving an automation rate of 2.5%. These results help ground discussions of AI automation in empirical evidence, setting a common basis for tracking AI impacts and enabling stakeholders to proactively navigate AI-driven labor automation.

comment: Website: https://www.remotelabor.ai

☆ LLMs Process Lists With General Filter Heads

Arnab Sen Sharma, Giordano Rogers, Natalie Shapira, David Bau

We investigate the mechanisms underlying a range of list-processing tasks in LLMs, and we find that LLMs have learned to encode a compact, causal representation of a general filtering operation that mirrors the generic "filter" function of functional programming. Using causal mediation analysis on a diverse set of list-processing tasks, we find that a small number of attention heads, which we dub filter heads, encode a compact representation of the filtering predicate in their query states at certain tokens. We demonstrate that this predicate representation is general and portable: it can be extracted and reapplied to execute the same filtering operation on different collections, presented in different formats, languages, or even in tasks. However, we also identify situations where transformer LMs can exploit a different strategy for filtering: eagerly evaluating if an item satisfies the predicate and storing this intermediate result as a flag directly in the item representations. Our results reveal that transformer LMs can develop human-interpretable implementations of abstract computational operations that generalize in ways that are surprisingly similar to strategies used in traditional functional programming patterns.

comment: Code and data at https://filter.baulab.info/

☆ Clone Deterministic 3D Worlds with Geometrically-Regularized World Models

Zaishuo Xia, Yukuan Lu, Xinyi Li, Yifan Xu, Yubei Chen

A world model is an internal model that simulates how the world evolves. Given past observations and actions, it predicts the future of both the embodied agent and its environment. Accurate world models are essential for enabling agents to think, plan, and reason effectively in complex, dynamic settings. Despite rapid progress, current world models remain brittle and degrade over long horizons. We argue that a central cause is representation quality: exteroceptive inputs (e.g., images) are high-dimensional, and lossy or entangled latents make dynamics learning unnecessarily hard. We therefore ask whether improving representation learning alone can substantially improve world-model performance. In this work, we take a step toward building a truly accurate world model by addressing a fundamental yet open problem: constructing a model that can fully clone and overfit to a deterministic 3D world. We propose Geometrically-Regularized World Models (GRWM), which enforces that consecutive points along a natural sensory trajectory remain close in latent representation space. This approach yields significantly improved latent representations that align closely with the true topology of the environment. GRWM is plug-and-play, requires only minimal architectural modification, scales with trajectory length, and is compatible with diverse latent generative backbones. Across deterministic 3D settings and long-horizon prediction tasks, GRWM significantly increases rollout fidelity and stability. Analyses show that its benefits stem from learning a latent manifold with superior geometric structure. These findings support a clear takeaway: improving representation learning is a direct and useful path to robust world models, delivering reliable long-horizon predictions without enlarging the dynamics module.

☆ Faithful and Fast Influence Function via Advanced Sampling

Jungyeon Koh, Hyeonsu Lyu, Jonggyu Jang, Hyun Jong Yang

How can we explain the influence of training data on black-box models? Influence functions (IFs) offer a post-hoc solution by utilizing gradients and Hessians. However, computing the Hessian for an entire dataset is resource-intensive, necessitating a feasible alternative. A common approach involves randomly sampling a small subset of the training data, but this method often results in highly inconsistent IF estimates due to the high variance in sample configurations. To address this, we propose two advanced sampling techniques based on features and logits. These samplers select a small yet representative subset of the entire dataset by considering the stochastic distribution of features or logits, thereby enhancing the accuracy of IF estimations. We validate our approach through class removal experiments, a typical application of IFs, using the F1-score to measure how effectively the model forgets the removed class while maintaining inference consistency on the remaining classes. Our method reduces computation time by 30.1% and memory usage by 42.2%, or improves the F1-score by 2.5% compared to the baseline.

☆ STaMP: Sequence Transformation and Mixed Precision for Low-Precision Activation Quantization

Marco Federici, Riccardo Del Chiaro, Boris van Breugel, Paul Whatmough, Markus Nagel

Quantization is the key method for reducing inference latency, power and memory footprint of generative AI models. However, accuracy often degrades sharply when activations are quantized below eight bits. Recent work suggests that invertible linear transformations (e.g. rotations) can aid quantization, by reparameterizing feature channels and weights. In this paper, we propose \textit{Sequence Transformation and Mixed Precision} (STaMP) quantization, a novel strategy that applies linear transformations along the \textit{sequence} dimension to exploit the strong local correlation in language and visual data. By keeping a small number of tokens in each intermediate activation at higher precision, we can maintain model accuracy at lower (average) activations bit-widths. We evaluate STaMP on recent LVM and LLM architectures, demonstrating that it significantly improves low bit width activation quantization and complements established activation and weight quantization methods including recent feature transformations.

comment: 10 pages main text, 8 pages supplementary material

☆ AMO-Bench: Large Language Models Still Struggle in High School Math Competitions

Shengnan An, Xunliang Cai, Xuezhi Cao, Xiaoyu Li, Yehao Lin, Junlin Liu, Xinxuan Lv, Dan Ma, Xuanlin Wang, Ziwen Wang, Shuang Zhou

We present AMO-Bench, an Advanced Mathematical reasoning benchmark with Olympiad level or even higher difficulty, comprising 50 human-crafted problems. Existing benchmarks have widely leveraged high school math competitions for evaluating mathematical reasoning capabilities of large language models (LLMs). However, many existing math competitions are becoming less effective for assessing top-tier LLMs due to performance saturation (e.g., AIME24/25). To address this, AMO-Bench introduces more rigorous challenges by ensuring all 50 problems are (1) cross-validated by experts to meet at least the International Mathematical Olympiad (IMO) difficulty standards, and (2) entirely original problems to prevent potential performance leakages from data memorization. Moreover, each problem in AMO-Bench requires only a final answer rather than a proof, enabling automatic and robust grading for evaluation. Experimental results across 26 LLMs on AMO-Bench show that even the best-performing model achieves only 52.4% accuracy on AMO-Bench, with most LLMs scoring below 40%. Beyond these poor performances, our further analysis reveals a promising scaling trend with increasing test-time compute on AMO-Bench. These results highlight the significant room for improving the mathematical reasoning in current LLMs. We release AMO-Bench to facilitate further research into advancing the reasoning abilities of language models. https://amo-bench.github.io/

comment: 14 pages, 9 figures

☆ The Oversight Game: Learning to Cooperatively Balance an AI Agent's Safety and Autonomy

William Overman, Mohsen Bayati

As increasingly capable agents are deployed, a central safety question is how to retain meaningful human control without modifying the underlying system. We study a minimal control interface where an agent chooses whether to act autonomously (play) or defer (ask), while a human simultaneously chooses whether to be permissive (trust) or to engage in oversight (oversee). If the agent defers, the human's choice determines the outcome, potentially leading to a corrective action or a system shutdown. We model this interaction as a two-player Markov Game. Our analysis focuses on cases where this game qualifies as a Markov Potential Game (MPG), a class of games where we can provide an alignment guarantee: under a structural assumption on the human's value function, any decision by the agent to act more autonomously that benefits itself cannot harm the human's value. We also analyze extensions to this MPG framework. Theoretically, this perspective provides conditions for a specific form of intrinsic alignment. If the reward structures of the human-agent game meet these conditions, we have a formal guarantee that the agent improving its own outcome will not harm the human's. Practically, this model motivates a transparent control layer with predictable incentives where the agent learns to defer when risky and act when safe, while its pretrained policy and the environment's reward structure remain untouched. Our gridworld simulation shows that through independent learning, the agent and human discover their optimal oversight roles. The agent learns to ask when uncertain and the human learns when to oversee, leading to an emergent collaboration that avoids safety violations introduced post-training. This demonstrates a practical method for making misaligned models safer after deployment.

☆ Deep sequence models tend to memorize geometrically; it is unclear why

Shahriar Noroozizadeh, Vaishnavh Nagarajan, Elan Rosenfeld, Sanjiv Kumar

In sequence modeling, the parametric memory of atomic facts has been predominantly abstracted as a brute-force lookup of co-occurrences between entities. We contrast this associative view against a geometric view of how memory is stored. We begin by isolating a clean and analyzable instance of Transformer reasoning that is incompatible with memory as strictly a storage of the local co-occurrences specified during training. Instead, the model must have somehow synthesized its own geometry of atomic facts, encoding global relationships between all entities, including non-co-occurring ones. This in turn has simplified a hard reasoning task involving an $\ell$-fold composition into an easy-to-learn 1-step geometric task. From this phenomenon, we extract fundamental aspects of neural embedding geometries that are hard to explain. We argue that the rise of such a geometry, despite optimizing over mere local associations, cannot be straightforwardly attributed to typical architectural or optimizational pressures. Counterintuitively, an elegant geometry is learned even when it is not more succinct than a brute-force lookup of associations. Then, by analyzing a connection to Node2Vec, we demonstrate how the geometry stems from a spectral bias that -- in contrast to prevailing theories -- indeed arises naturally despite the lack of various pressures. This analysis also points to practitioners a visible headroom to make Transformer memory more strongly geometric. We hope the geometric view of parametric memory encourages revisiting the default intuitions that guide researchers in areas like knowledge acquisition, capacity, discovery and unlearning.

☆ A General Incentives-Based Framework for Fairness in Multi-agent Resource Allocation

Ashwin Kumar, William Yeoh

We introduce the General Incentives-based Framework for Fairness (GIFF), a novel approach for fair multi-agent resource allocation that infers fair decision-making from standard value functions. In resource-constrained settings, agents optimizing for efficiency often create inequitable outcomes. Our approach leverages the action-value (Q-)function to balance efficiency and fairness without requiring additional training. Specifically, our method computes a local fairness gain for each action and introduces a counterfactual advantage correction term to discourage over-allocation to already well-off agents. This approach is formalized within a centralized control setting, where an arbitrator uses the GIFF-modified Q-values to solve an allocation problem. Empirical evaluations across diverse domains, including dynamic ridesharing, homelessness prevention, and a complex job allocation task-demonstrate that our framework consistently outperforms strong baselines and can discover far-sighted, equitable policies. The framework's effectiveness is supported by a theoretical foundation; we prove its fairness surrogate is a principled lower bound on the true fairness improvement and that its trade-off parameter offers monotonic tuning. Our findings establish GIFF as a robust and principled framework for leveraging standard reinforcement learning components to achieve more equitable outcomes in complex multi-agent systems.

☆ Cross-Platform Evaluation of Reasoning Capabilities in Foundation Models

J. de Curtò, I. de Zarzà, Pablo García, Jordi Cabot

This paper presents a comprehensive cross-platform evaluation of reasoning capabilities in contemporary foundation models, establishing an infrastructure-agnostic benchmark across three computational paradigms: HPC supercomputing (MareNostrum 5), cloud platforms (Nebius AI Studio), and university clusters (a node with eight H200 GPUs). We evaluate 15 foundation models across 79 problems spanning eight academic domains (Physics, Mathematics, Chemistry, Economics, Biology, Statistics, Calculus, and Optimization) through three experimental phases: (1) Baseline establishment: Six models (Mixtral-8x7B, Phi-3, LLaMA 3.1-8B, Gemma-2-9b, Mistral-7B, OLMo-7B) evaluated on 19 problems using MareNostrum 5, establishing methodology and reference performance; (2) Infrastructure validation: The 19-problem benchmark repeated on university cluster (seven models including Falcon-Mamba state-space architecture) and Nebius AI Studio (nine state-of-the-art models: Hermes-4 70B/405B, LLaMA 3.1-405B/3.3-70B, Qwen3 30B/235B, DeepSeek-R1, GPT-OSS 20B/120B) to confirm infrastructure-agnostic reproducibility; (3) Extended evaluation: Full 79-problem assessment on both university cluster and Nebius platforms, probing generalization at scale across architectural diversity. The findings challenge conventional scaling assumptions, establish training data quality as more critical than model size, and provide actionable guidelines for model selection across educational, production, and research contexts. The tri-infrastructure methodology and 79-problem benchmark enable longitudinal tracking of reasoning capabilities as foundation models evolve.

☆ ExpertFlow: Adaptive Expert Scheduling and Memory Coordination for Efficient MoE Inference

Zixu Shen, Kexin Chu, Yifan Zhang, Dawei Xiang, Runxin Wu, Wei Zhang

The expansion of large language models is increasingly limited by the constrained memory capacity of modern GPUs. To mitigate this, Mixture-of-Experts (MoE) architectures activate only a small portion of parameters during inference, significantly lowering both memory demand and computational overhead. However, conventional MoE inference approaches, which select active experts independently at each layer, often introduce considerable latency because of frequent parameter transfers between host and GPU memory. In addition, current cross-layer prediction strategies, which are typically based on fixed steps, lack adaptability across different hardware platforms and workloads, thereby reducing their robustness and effectiveness. To address these challenges, we present ExpertFlow, a runtime system for MoE inference that combines adaptive expert prefetching and cache-aware routing. ExpertFlow continuously adjusts its prediction horizon for expert activation by leveraging runtime statistics such as transfer bandwidth, parameter dimensionality, and model feedback signals. Furthermore, it incorporates a hybrid cross-layer prediction scheme that fuses pregating information with intermediate computational states to anticipate future expert needs. By adaptively refining prefetching decisions and aligning them with actual usage behavior, ExpertFlow effectively decreases cache misses and removes latency caused by expert swap-ins. Our evaluation demonstrates that ExpertFlow reduces model stall time to less than 0.1% of the baseline, highlighting its capability to optimize MoE inference under stringent memory constraints.

comment: 12 pages, 11 figures

☆ Non-Convex Over-the-Air Heterogeneous Federated Learning: A Bias-Variance Trade-off

Muhammad Faraz Ul Abrar, Nicolò Michelusi

Over-the-air (OTA) federated learning (FL) has been well recognized as a scalable paradigm that exploits the waveform superposition of the wireless multiple-access channel to aggregate model updates in a single use. Existing OTA-FL designs largely enforce zero-bias model updates by either assuming \emph{homogeneous} wireless conditions (equal path loss across devices) or forcing zero-bias updates to guarantee convergence. Under \emph{heterogeneous} wireless scenarios, however, such designs are constrained by the weakest device and inflate the update variance. Moreover, prior analyses of biased OTA-FL largely address convex objectives, while most modern AI models are highly non-convex. Motivated by these gaps, we study OTA-FL with stochastic gradient descent (SGD) for general smooth non-convex objectives under wireless heterogeneity. We develop novel OTA-FL SGD updates that allow a structured, time-invariant model bias while facilitating reduced variance updates. We derive a finite-time stationarity bound (expected time average squared gradient norm) that explicitly reveals a bias-variance trade-off. To optimize this trade-off, we pose a non-convex joint OTA power-control design and develop an efficient successive convex approximation (SCA) algorithm that requires only statistical CSI at the base station. Experiments on a non-convex image classification task validate the approach: the SCA-based design accelerates convergence via an optimized bias and improves generalization over prior OTA-FL baselines.

☆ Unveiling Intrinsic Text Bias in Multimodal Large Language Models through Attention Key-Space Analysis

Xinhan Zheng, Huyu Wu, Xueting Wang, Haiyun Jiang

Multimodal large language models (MLLMs) exhibit a pronounced preference for textual inputs when processing vision-language data, limiting their ability to reason effectively from visual evidence. Unlike prior studies that attribute this text bias to external factors such as data imbalance or instruction tuning, we propose that the bias originates from the model's internal architecture. Specifically, we hypothesize that visual key vectors (Visual Keys) are out-of-distribution (OOD) relative to the text key space learned during language-only pretraining. Consequently, these visual keys receive systematically lower similarity scores during attention computation, leading to their under-utilization in the context representation. To validate this hypothesis, we extract key vectors from LLaVA and Qwen2.5-VL and analyze their distributional structures using qualitative (t-SNE) and quantitative (Jensen-Shannon divergence) methods. The results provide direct evidence that visual and textual keys occupy markedly distinct subspaces within the attention space. The inter-modal divergence is statistically significant, exceeding intra-modal variation by several orders of magnitude. These findings reveal that text bias arises from an intrinsic misalignment within the attention key space rather than solely from external data factors.

☆ On the limitation of evaluating machine unlearning using only a single training seed

Jamie Lanyon, Axel Finke, Petros Andreou, Georgina Cosma

Machine unlearning (MU) aims to remove the influence of certain data points from a trained model without costly retraining. Most practical MU algorithms are only approximate and their performance can only be assessed empirically. Care must therefore be taken to make empirical comparisons as representative as possible. A common practice is to run the MU algorithm multiple times independently starting from the same trained model. In this work, we demonstrate that this practice can give highly non-representative results because -- even for the same architecture and same dataset -- some MU methods can be highly sensitive to the choice of random number seed used for model training. We therefore recommend that empirical comphttps://info.arxiv.org/help/prep#commentsarisons of MU algorithms should also reflect the variability across different model training seeds.

comment: mini paper, 2 figures

☆ Delegated Authorization for Agents Constrained to Semantic Task-to-Scope Matching

Majed El Helou, Chiara Troiani, Benjamin Ryder, Jean Diaconu, Hervé Muyal, Marcelo Yannuzzi

Authorizing Large Language Model driven agents to dynamically invoke tools and access protected resources introduces significant risks, since current methods for delegating authorization grant overly broad permissions and give access to tools allowing agents to operate beyond the intended task scope. We introduce and assess a delegated authorization model enabling authorization servers to semantically inspect access requests to protected resources, and issue access tokens constrained to the minimal set of scopes necessary for the agents' assigned tasks. Given the unavailability of datasets centered on delegated authorization flows, particularly including both semantically appropriate and inappropriate scope requests for a given task, we introduce ASTRA, a dataset and data generation pipeline for benchmarking semantic matching between tasks and scopes. Our experiments show both the potential and current limitations of model-based matching, particularly as the number of scopes needed for task completion increases. Our results highlight the need for further research into semantic matching techniques enabling intent-aware authorization for multi-agent and tool-augmented applications, including fine-grained control, such as Task-Based Access Control (TBAC).

comment: Paper page at https://outshift-open.github.io/ASTRA

☆ The End of Manual Decoding: Towards Truly End-to-End Language Models

Zhichao Wang, Dongyang Ma, Xinting Huang, Deng Cai, Tian Lan, Jiahao Xu, Haitao Mi, Xiaoying Tang, Yan Wang

The "end-to-end" label for LLMs is a misnomer. In practice, they depend on a non-differentiable decoding process that requires laborious, hand-tuning of hyperparameters like temperature and top-p. This paper introduces AutoDeco, a novel architecture that enables truly "end-to-end" generation by learning to control its own decoding strategy. We augment the standard transformer with lightweight heads that, at each step, dynamically predict context-specific temperature and top-p values alongside the next-token logits. This approach transforms decoding into a parametric, token-level process, allowing the model to self-regulate its sampling strategy within a single forward pass. Through extensive experiments on eight benchmarks, we demonstrate that AutoDeco not only significantly outperforms default decoding strategies but also achieves performance comparable to an oracle-tuned baseline derived from "hacking the test set"-a practical upper bound for any static method. Crucially, we uncover an emergent capability for instruction-based decoding control: the model learns to interpret natural language commands (e.g., "generate with low randomness") and adjusts its predicted temperature and top-p on a token-by-token basis, opening a new paradigm for steerable and interactive LLM decoding.

☆ Process Integrated Computer Vision for Real-Time Failure Prediction in Steel Rolling Mill

Vaibhav Kurrey, Sivakalyan Pujari, Gagan Raj Gupta

We present a long-term deployment study of a machine vision-based anomaly detection system for failure prediction in a steel rolling mill. The system integrates industrial cameras to monitor equipment operation, alignment, and hot bar motion in real time along the process line. Live video streams are processed on a centralized video server using deep learning models, enabling early prediction of equipment failures and process interruptions, thereby reducing unplanned breakdown costs. Server-based inference minimizes the computational load on industrial process control systems (PLCs), supporting scalable deployment across production lines with minimal additional resources. By jointly analyzing sensor data from data acquisition systems and visual inputs, the system identifies the location and probable root causes of failures, providing actionable insights for proactive maintenance. This integrated approach enhances operational reliability, productivity, and profitability in industrial manufacturing environments.

☆ Evontree: Ontology Rule-Guided Self-Evolution of Large Language Models

Mingchen Tu, Zhiqiang Liu, Juan Li, Liangyurui Liu, Junjie Wang, Lei Liang, Wen Zhang

Large language models (LLMs) have demonstrated exceptional capabilities across multiple domains by leveraging massive pre-training and curated fine-tuning data. However, in data-sensitive fields such as healthcare, the lack of high-quality, domain-specific training corpus hinders LLMs' adaptation for specialized applications. Meanwhile, domain experts have distilled domain wisdom into ontology rules, which formalize relationships among concepts and ensure the integrity of knowledge management repositories. Viewing LLMs as implicit repositories of human knowledge, we propose Evontree, a novel framework that leverages a small set of high-quality ontology rules to systematically extract, validate, and enhance domain knowledge within LLMs, without requiring extensive external datasets. Specifically, Evontree extracts domain ontology from raw models, detects inconsistencies using two core ontology rules, and reinforces the refined knowledge via self-distilled fine-tuning. Extensive experiments on medical QA benchmarks with Llama3-8B-Instruct and Med42-v2 demonstrate consistent outperformance over both unmodified models and leading supervised baselines, achieving up to a 3.7% improvement in accuracy. These results confirm the effectiveness, efficiency, and robustness of our approach for low-resource domain adaptation of LLMs.

☆ The Era of Agentic Organization: Learning to Organize with Language Models

Zewen Chi, Li Dong, Qingxiu Dong, Yaru Hao, Xun Wu, Shaohan Huang, Furu Wei

We envision a new era of AI, termed agentic organization, where agents solve complex problems by working collaboratively and concurrently, enabling outcomes beyond individual intelligence. To realize this vision, we introduce asynchronous thinking (AsyncThink) as a new paradigm of reasoning with large language models, which organizes the internal thinking process into concurrently executable structures. Specifically, we propose a thinking protocol where an organizer dynamically assigns sub-queries to workers, merges intermediate knowledge, and produces coherent solutions. More importantly, the thinking structure in this protocol can be further optimized through reinforcement learning. Experiments demonstrate that AsyncThink achieves 28% lower inference latency compared to parallel thinking while improving accuracy on mathematical reasoning. Moreover, AsyncThink generalizes its learned asynchronous thinking capabilities, effectively tackling unseen tasks without additional training.

☆ Hybrid DQN-TD3 Reinforcement Learning for Autonomous Navigation in Dynamic Environments

Xiaoyi He, Danggui Chen, Zhenshuo Zhang, Zimeng Bai

This paper presents a hierarchical path-planning and control framework that combines a high-level Deep Q-Network (DQN) for discrete sub-goal selection with a low-level Twin Delayed Deep Deterministic Policy Gradient (TD3) controller for continuous actuation. The high-level module selects behaviors and sub-goals; the low-level module executes smooth velocity commands. We design a practical reward shaping scheme (direction, distance, obstacle avoidance, action smoothness, collision penalty, time penalty, and progress), together with a LiDAR-based safety gate that prevents unsafe motions. The system is implemented in ROS + Gazebo (TurtleBot3) and evaluated with PathBench metrics, including success rate, collision rate, path efficiency, and re-planning efficiency, in dynamic and partially observable environments. Experiments show improved success rate and sample efficiency over single-algorithm baselines (DQN or TD3 alone) and rule-based planners, with better generalization to unseen obstacle configurations and reduced abrupt control changes. Code and evaluation scripts are available at the project repository.

comment: 6 pages, 5 figures; ROS+Gazebo (TurtleBot3) implementation; evaluation with PathBench metrics; code (primary): https://github.com/MayaCHEN-github/HierarchicalRL-robot-navigation; mirror (for reproducibility): https://github.com/ShowyHe/DRL-robot-navigation

☆ Aeolus: A Multi-structural Flight Delay Dataset

Lin Xu, Xinyun Yuan, Yuxuan Liang, Suwan Yin, Yuankai Wu

We introduce Aeolus, a large-scale Multi-modal Flight Delay Dataset designed to advance research on flight delay prediction and support the development of foundation models for tabular data. Existing datasets in this domain are typically limited to flat tabular structures and fail to capture the spatiotemporal dynamics inherent in delay propagation. Aeolus addresses this limitation by providing three aligned modalities: (i) a tabular dataset with rich operational, meteorological, and airportlevel features for over 50 million flights; (ii) a flight chain module that models delay propagation along sequential flight legs, capturing upstream and downstream dependencies; and (iii) a flight network graph that encodes shared aircraft, crew, and airport resource connections, enabling cross-flight relational reasoning. The dataset is carefully constructed with temporal splits, comprehensive features, and strict leakage prevention to support realistic and reproducible machine learning evaluation. Aeolus supports a broad range of tasks, including regression, classification, temporal structure modeling, and graph learning, serving as a unified benchmark across tabular, sequential, and graph modalities. We release baseline experiments and preprocessing tools to facilitate adoption. Aeolus fills a key gap for both domain-specific modeling and general-purpose structured data research.Our source code and data can be accessed at https://github.com/Flnny/Delay-data

☆ Normative Reasoning in Large Language Models: A Comparative Benchmark from Logical and Modal Perspectives EMNLP 2025

Kentaro Ozeki, Risako Ando, Takanobu Morishita, Hirohiko Abe, Koji Mineshima, Mitsuhiro Okada

Normative reasoning is a type of reasoning that involves normative or deontic modality, such as obligation and permission. While large language models (LLMs) have demonstrated remarkable performance across various reasoning tasks, their ability to handle normative reasoning remains underexplored. In this paper, we systematically evaluate LLMs' reasoning capabilities in the normative domain from both logical and modal perspectives. Specifically, to assess how well LLMs reason with normative modals, we make a comparison between their reasoning with normative modals and their reasoning with epistemic modals, which share a common formal structure. To this end, we introduce a new dataset covering a wide range of formal patterns of reasoning in both normative and epistemic domains, while also incorporating non-formal cognitive factors that influence human reasoning. Our results indicate that, although LLMs generally adhere to valid reasoning patterns, they exhibit notable inconsistencies in specific types of normative reasoning and display cognitive biases similar to those observed in psychological studies of human reasoning. These findings highlight challenges in achieving logical consistency in LLMs' normative reasoning and provide insights for enhancing their reliability. All data and code are released publicly at https://github.com/kmineshima/NeuBAROCO.

comment: Accepted to the 8th BlackboxNLP Workshop at EMNLP 2025

☆ Agentic AI Home Energy Management System: A Large Language Model Framework for Residential Load Scheduling

Reda El Makroum, Sebastian Zwickl-Bernhard, Lukas Kranzl

The electricity sector transition requires substantial increases in residential demand response capacity, yet Home Energy Management Systems (HEMS) adoption remains limited by user interaction barriers requiring translation of everyday preferences into technical parameters. While large language models have been applied to energy systems as code generators and parameter extractors, no existing implementation deploys LLMs as autonomous coordinators managing the complete workflow from natural language input to multi-appliance scheduling. This paper presents an agentic AI HEMS where LLMs autonomously coordinate multi-appliance scheduling from natural language requests to device control, achieving optimal scheduling without example demonstrations. A hierarchical architecture combining one orchestrator with three specialist agents uses the ReAct pattern for iterative reasoning, enabling dynamic coordination without hardcoded workflows while integrating Google Calendar for context-aware deadline extraction. Evaluation across three open-source models using real Austrian day-ahead electricity prices reveals substantial capability differences. Llama-3.3-70B successfully coordinates all appliances across all scenarios to match cost-optimal benchmarks computed via mixed-integer linear programming, while other models achieve perfect single-appliance performance but struggle to coordinate all appliances simultaneously. Progressive prompt engineering experiments demonstrate that analytical query handling without explicit guidance remains unreliable despite models' general reasoning capabilities. We open-source the complete system including orchestration logic, agent prompts, tools, and web interfaces to enable reproducibility, extension, and future research.

comment: 34 pages, 9 figures. Code available at https://github.com/RedaElMakroum/agentic-ai-hems

☆ ResMatching: Noise-Resilient Computational Super-Resolution via Guided Conditional Flow Matching

Anirban Ray, Vera Galinova, Florian Jug

Computational Super-Resolution (CSR) in fluorescence microscopy has, despite being an ill-posed problem, a long history. At its very core, CSR is about finding a prior that can be used to extrapolate frequencies in a micrograph that have never been imaged by the image-generating microscope. It stands to reason that, with the advent of better data-driven machine learning techniques, stronger prior can be learned and hence CSR can lead to better results. Here, we present ResMatching, a novel CSR method that uses guided conditional flow matching to learn such improved data-priors. We evaluate ResMatching on 4 diverse biological structures from the BioSR dataset and compare its results against 7 baselines. ResMatching consistently achieves competitive results, demonstrating in all cases the best trade-off between data fidelity and perceptual realism. We observe that CSR using ResMatching is particularly effective in cases where a strong prior is hard to learn, e.g. when the given low-resolution images contain a lot of noise. Additionally, we show that ResMatching can be used to sample from an implicitly learned posterior distribution and that this distribution is calibrated for all tested use-cases, enabling our method to deliver a pixel-wise data-uncertainty term that can guide future users to reject uncertain predictions.

comment: 5 pages, 4 figures

☆ Stop Wasting Your Tokens: Towards Efficient Runtime Multi-Agent Systems

Fulin Lin, Shaowen Chen, Ruishan Fang, Hongwei Wang, Tao Lin

While Multi-Agent Systems (MAS) excel at complex tasks, their growing autonomy with operational complexity often leads to critical inefficiencies, such as excessive token consumption and failures arising from misinformation. Existing methods primarily focus on post-hoc failure attribution, lacking proactive, real-time interventions to enhance robustness and efficiency. To this end, we introduce SupervisorAgent, a lightweight and modular framework for runtime, adaptive supervision that operates without altering the base agent's architecture. Triggered by an LLM-free adaptive filter, SupervisorAgent intervenes at critical junctures to proactively correct errors, guide inefficient behaviors, and purify observations. On the challenging GAIA benchmark, SupervisorAgent reduces the token consumption of the Smolagent framework by an average of 29.45% without compromising its success rate. Extensive experiments across five additional benchmarks (math reasoning, code generation, and question answering) and various SoTA foundation models validate the broad applicability and robustness of our approach. The code is available at https://github.com/LINs-lab/SupervisorAgent.

☆ InfoFlow: Reinforcing Search Agent Via Reward Density Optimization

Kun Luo, Hongjin Qian, Zheng Liu, Ziyi Xia, Shitao Xiao, Siqi Bao, Jun Zhao, Kang Liu

Reinforcement Learning with Verifiable Rewards (RLVR) is a promising approach for enhancing agentic deep search. However, its application is often hindered by low \textbf{Reward Density} in deep search scenarios, where agents expend significant exploratory costs for infrequent and often null final rewards. In this paper, we formalize this challenge as the \textbf{Reward Density Optimization} problem, which aims to improve the reward obtained per unit of exploration cost. This paper introduce \textbf{InfoFlow}, a systematic framework that tackles this problem from three aspects. 1) \textbf{Subproblem decomposition}: breaking down long-range tasks to assign process rewards, thereby providing denser learning signals. 2) \textbf{Failure-guided hints}: injecting corrective guidance into stalled trajectories to increase the probability of successful outcomes. 3) \textbf{Dual-agent refinement}: employing a dual-agent architecture to offload the cognitive burden of deep exploration. A refiner agent synthesizes the search history, which effectively compresses the researcher's perceived trajectory, thereby reducing exploration cost and increasing the overall reward density. We evaluate InfoFlow on multiple agentic search benchmarks, where it significantly outperforms strong baselines, enabling lightweight LLMs to achieve performance comparable to advanced proprietary LLMs.

☆ Multiclass Local Calibration With the Jensen-Shannon Distance

Cesare Barbera, Lorenzo Perini, Giovanni De Toni, Andrea Passerini, Andrea Pugnana

Developing trustworthy Machine Learning (ML) models requires their predicted probabilities to be well-calibrated, meaning they should reflect true-class frequencies. Among calibration notions in multiclass classification, strong calibration is the most stringent, as it requires all predicted probabilities to be simultaneously calibrated across all classes. However, existing approaches to multiclass calibration lack a notion of distance among inputs, which makes them vulnerable to proximity bias: predictions in sparse regions of the feature space are systematically miscalibrated. This is especially relevant in high-stakes settings, such as healthcare, where the sparse instances are exactly those most at risk of biased treatment. In this work, we address this main shortcoming by introducing a local perspective on multiclass calibration. First, we formally define multiclass local calibration and establish its relationship with strong calibration. Second, we theoretically analyze the pitfalls of existing evaluation metrics when applied to multiclass local calibration. Third, we propose a practical method for enhancing local calibration in Neural Networks, which enforces alignment between predicted probabilities and local estimates of class frequencies using the Jensen-Shannon distance. Finally, we empirically validate our approach against existing multiclass calibration techniques.

☆ Adaptive Inverse Kinematics Framework for Learning Variable-Length Tool Manipulation in Robotics

Prathamesh Kothavale, Sravani Boddepalli

Conventional robots possess a limited understanding of their kinematics and are confined to preprogrammed tasks, hindering their ability to leverage tools efficiently. Driven by the essential components of tool usage - grasping the desired outcome, selecting the most suitable tool, determining optimal tool orientation, and executing precise manipulations - we introduce a pioneering framework. Our novel approach expands the capabilities of the robot's inverse kinematics solver, empowering it to acquire a sequential repertoire of actions using tools of varying lengths. By integrating a simulation-learned action trajectory with the tool, we showcase the practicality of transferring acquired skills from simulation to real-world scenarios through comprehensive experimentation. Remarkably, our extended inverse kinematics solver demonstrates an impressive error rate of less than 1 cm. Furthermore, our trained policy achieves a mean error of 8 cm in simulation. Noteworthy, our model achieves virtually indistinguishable performance when employing two distinct tools of different lengths. This research provides an indication of potential advances in the exploration of all four fundamental aspects of tool usage, enabling robots to master the intricate art of tool manipulation across diverse tasks.

comment: 10 pages, 5 figures. Demonstrates a reinforcement learning framework for adaptive tool manipulation with variable-length extensions

☆ EdgeRunner 20B: Military Task Parity with GPT-5 while Running on the Edge

Jack FitzGerald, Aristotelis Lazaridis, Dylan Bates, Aman Sharma, Jonnathan Castillo, Yousif Azami, Sean Bailey, Jeremy Cao, Peter Damianov, Kevin de Haan, Luke Kerbs, Vincent Lu, Joseph Madigan, Jeremy McLaurin, Jonathan Tainer, Dave Anderson, Jonathan Beck, Jamie Cuticello, Colton Malkerson, Tyler Saltsman

We present EdgeRunner 20B, a fine-tuned version of gpt-oss-20b optimized for military tasks. EdgeRunner 20B was trained on 1.6M high-quality records curated from military documentation and websites. We also present four new tests sets: (a) combat arms, (b) combat medic, (c) cyber operations, and (d) mil-bench-5k (general military knowledge). On these military test sets, EdgeRunner 20B matches or exceeds GPT-5 task performance with 95%+ statistical significance, except for the high reasoning setting on the combat medic test set and the low reasoning setting on the mil-bench-5k test set. Versus gpt-oss-20b, there is no statistically-significant regression on general-purpose benchmarks like ARC-C, GPQA Diamond, GSM8k, IFEval, MMLU Pro, or TruthfulQA, except for GSM8k in the low reasoning setting. We also present analyses on hyperparameter settings, cost, and throughput. These findings show that small, locally-hosted models are ideal solutions for data-sensitive operations such as in the military domain, allowing for deployment in air-gapped edge devices.

comment: 19 pages

☆ The Structure of Relation Decoding Linear Operators in Large Language Models NeurIPS 2025

Miranda Anna Christ, Adrián Csiszárik, Gergely Becsó, Dániel Varga

This paper investigates the structure of linear operators introduced in Hernandez et al. [2023] that decode specific relational facts in transformer language models. We extend their single-relation findings to a collection of relations and systematically chart their organization. We show that such collections of relation decoders can be highly compressed by simple order-3 tensor networks without significant loss in decoding accuracy. To explain this surprising redundancy, we develop a cross-evaluation protocol, in which we apply each linear decoder operator to the subjects of every other relation. Our results reveal that these linear maps do not encode distinct relations, but extract recurring, coarse-grained semantic properties (e.g., country of capital city and country of food are both in the country-of-X property). This property-centric structure clarifies both the operators' compressibility and highlights why they generalize only to new relations that are semantically close. Our findings thus interpret linear relational decoding in transformer language models as primarily property-based, rather than relation-specific.

comment: NeurIPS 2025 (Spotlight)

☆ Human-AI Complementarity: A Goal for Amplified Oversight

Rishub Jain, Sophie Bridgers, Lili Janzer, Rory Greig, Tian Huey Teh, Vladimir Mikulik

Human feedback is critical for aligning AI systems to human values. As AI capabilities improve and AI is used to tackle more challenging tasks, verifying quality and safety becomes increasingly challenging. This paper explores how we can leverage AI to improve the quality of human oversight. We focus on an important safety problem that is already challenging for humans: fact-verification of AI outputs. We find that combining AI ratings and human ratings based on AI rater confidence is better than relying on either alone. Giving humans an AI fact-verification assistant further improves their accuracy, but the type of assistance matters. Displaying AI explanation, confidence, and labels leads to over-reliance, but just showing search results and evidence fosters more appropriate trust. These results have implications for Amplified Oversight -- the challenge of combining humans and AI to supervise AI systems even as they surpass human expert performance.

☆ Inside CORE-KG: Evaluating Structured Prompting and Coreference Resolution for Knowledge Graphs ICDM 2025

Dipak Meher, Carlotta Domeniconi

Human smuggling networks are increasingly adaptive and difficult to analyze. Legal case documents offer critical insights but are often unstructured, lexically dense, and filled with ambiguous or shifting references, which pose significant challenges for automated knowledge graph (KG) construction. While recent LLM-based approaches improve over static templates, they still generate noisy, fragmented graphs with duplicate nodes due to the absence of guided extraction and coreference resolution. The recently proposed CORE-KG framework addresses these limitations by integrating a type-aware coreference module and domain-guided structured prompts, significantly reducing node duplication and legal noise. In this work, we present a systematic ablation study of CORE-KG to quantify the individual contributions of its two key components. Our results show that removing coreference resolution results in a 28.32% increase in node duplication and a 4.32% increase in noisy nodes, while removing structured prompts leads to a 4.34% increase in node duplication and a 73.33% increase in noisy nodes. These findings offer empirical insights for designing robust LLM-based pipelines for extracting structured representations from complex legal texts.

comment: ICDM 2025 Workshop

☆ Simulating and Experimenting with Social Media Mobilization Using LLM Agents

Sadegh Shirani, Mohsen Bayati

Online social networks have transformed the ways in which political mobilization messages are disseminated, raising new questions about how peer influence operates at scale. Building on the landmark 61-million-person Facebook experiment \citep{bond201261}, we develop an agent-based simulation framework that integrates real U.S. Census demographic distributions, authentic Twitter network topology, and heterogeneous large language model (LLM) agents to examine the effect of mobilization messages on voter turnout. Each simulated agent is assigned demographic attributes, a personal political stance, and an LLM variant (\texttt{GPT-4.1}, \texttt{GPT-4.1-Mini}, or \texttt{GPT-4.1-Nano}) reflecting its political sophistication. Agents interact over realistic social network structures, receiving personalized feeds and dynamically updating their engagement behaviors and voting intentions. Experimental conditions replicate the informational and social mobilization treatments of the original Facebook study. Across scenarios, the simulator reproduces qualitative patterns observed in field experiments, including stronger mobilization effects under social message treatments and measurable peer spillovers. Our framework provides a controlled, reproducible environment for testing counterfactual designs and sensitivity analyses in political mobilization research, offering a bridge between high-validity field experiments and flexible computational modeling.\footnote{Code and data available at https://github.com/CausalMP/LLM-SocioPol}

☆ Context Engineering 2.0: The Context of Context Engineering

Qishuo Hua, Lyumanshan Ye, Dayuan Fu, Yang Xiao, Xiaojie Cai, Yunze Wu, Jifan Lin, Junfei Wang, Pengfei Liu

Karl Marx once wrote that ``the human essence is the ensemble of social relations'', suggesting that individuals are not isolated entities but are fundamentally shaped by their interactions with other entities, within which contexts play a constitutive and essential role. With the advent of computers and artificial intelligence, these contexts are no longer limited to purely human--human interactions: human--machine interactions are included as well. Then a central question emerges: How can machines better understand our situations and purposes? To address this challenge, researchers have recently introduced the concept of context engineering. Although it is often regarded as a recent innovation of the agent era, we argue that related practices can be traced back more than twenty years. Since the early 1990s, the field has evolved through distinct historical phases, each shaped by the intelligence level of machines: from early human--computer interaction frameworks built around primitive computers, to today's human--agent interaction paradigms driven by intelligent agents, and potentially to human--level or superhuman intelligence in the future. In this paper, we situate context engineering, provide a systematic definition, outline its historical and conceptual landscape, and examine key design considerations for practice. By addressing these questions, we aim to offer a conceptual foundation for context engineering and sketch its promising future. This paper is a stepping stone for a broader community effort toward systematic context engineering in AI systems.

☆ LINK-KG: LLM-Driven Coreference-Resolved Knowledge Graphs for Human Smuggling Networks

Dipak Meher, Carlotta Domeniconi, Guadalupe Correa-Cabrera

Human smuggling networks are complex and constantly evolving, making them difficult to analyze comprehensively. Legal case documents offer rich factual and procedural insights into these networks but are often long, unstructured, and filled with ambiguous or shifting references, posing significant challenges for automated knowledge graph (KG) construction. Existing methods either overlook coreference resolution or fail to scale beyond short text spans, leading to fragmented graphs and inconsistent entity linking. We propose LINK-KG, a modular framework that integrates a three-stage, LLM-guided coreference resolution pipeline with downstream KG extraction. At the core of our approach is a type-specific Prompt Cache, which consistently tracks and resolves references across document chunks, enabling clean and disambiguated narratives for structured knowledge graph construction from both short and long legal texts. LINK-KG reduces average node duplication by 45.21% and noisy nodes by 32.22% compared to baseline methods, resulting in cleaner and more coherent graph structures. These improvements establish LINK-KG as a strong foundation for analyzing complex criminal networks.

comment: Accepted in ICKG 2025 Conference, 8 Pages, 2 Figures

☆ Bayesian Network Fusion of Large Language Models for Sentiment Analysis

Rasoul Amirzadeh, Dhananjay Thiruvady, Fatemeh Shiri

Large language models (LLMs) continue to advance, with an increasing number of domain-specific variants tailored for specialised tasks. However, these models often lack transparency and explainability, can be costly to fine-tune, require substantial prompt engineering, yield inconsistent results across domains, and impose significant adverse environmental impact due to their high computational demands. To address these challenges, we propose the Bayesian network LLM fusion (BNLF) framework, which integrates predictions from three LLMs, including FinBERT, RoBERTa, and BERTweet, through a probabilistic mechanism for sentiment analysis. BNLF performs late fusion by modelling the sentiment predictions from multiple LLMs as probabilistic nodes within a Bayesian network. Evaluated across three human-annotated financial corpora with distinct linguistic and contextual characteristics, BNLF demonstrates consistent gains of about six percent in accuracy over the baseline LLMs, underscoring its robustness to dataset variability and the effectiveness of probabilistic fusion for interpretable sentiment classification.

☆ Who Has The Final Say? Conformity Dynamics in ChatGPT's Selections

Clarissa Sabrina Arlinghaus, Tristan Kenneweg, Barbara Hammer, Günter W. Maier

Large language models (LLMs) such as ChatGPT are increasingly integrated into high-stakes decision-making, yet little is known about their susceptibility to social influence. We conducted three preregistered conformity experiments with GPT-4o in a hiring context. In a baseline study, GPT consistently favored the same candidate (Profile C), reported moderate expertise (M = 3.01) and high certainty (M = 3.89), and rarely changed its choice. In Study 1 (GPT + 8), GPT faced unanimous opposition from eight simulated partners and almost always conformed (99.9%), reporting lower certainty and significantly elevated self-reported informational and normative conformity (p < .001). In Study 2 (GPT + 1), GPT interacted with a single partner and still conformed in 40.2% of disagreement trials, reporting less certainty and more normative conformity. Across studies, results demonstrate that GPT does not act as an independent observer but adapts to perceived social consensus. These findings highlight risks of treating LLMs as neutral decision aids and underline the need to elicit AI judgments prior to exposing them to human opinions.

comment: 5 pages, 5 figures, HAI 2025: Workshop on Socially Aware and Cooperative Intelligent Systems

☆ Counteracting Matthew Effect in Self-Improvement of LVLMs through Head-Tail Re-balancing

Xin Guo, Zhiheng Xi, Yiwen Ding, Yitao Zhai, Xiaowei Shi, Xunliang Cai, Tao Gui, Qi Zhang, Xuanjing Huang

Self-improvement has emerged as a mainstream paradigm for advancing the reasoning capabilities of large vision-language models (LVLMs), where models explore and learn from successful trajectories iteratively. However, we identify a critical issue during this process: the model excels at generating high-quality trajectories for simple queries (i.e., head data) but struggles with more complex ones (i.e., tail data). This leads to an imbalanced optimization that drives the model to prioritize simple reasoning skills, while hindering its ability to tackle more complex reasoning tasks. Over iterations, this imbalance becomes increasingly pronounced--a dynamic we term the "Matthew effect"--which ultimately hinders further model improvement and leads to performance bottlenecks. To counteract this challenge, we introduce four efficient strategies from two perspectives: distribution-reshaping and trajectory-resampling, to achieve head-tail re-balancing during the exploration-and-learning self-improvement process. Extensive experiments on Qwen2-VL-7B-Instruct and InternVL2.5-4B models across visual reasoning tasks demonstrate that our methods consistently improve visual reasoning capabilities, outperforming vanilla self-improvement by 3.86 points on average.

comment: Preprint

☆ SecureReviewer: Enhancing Large Language Models for Secure Code Review through Secure-aware Fine-tuning ICSE 2026

Fang Liu, Simiao Liu, Yinghao Zhu, Xiaoli Lian, Li Zhang

Identifying and addressing security issues during the early phase of the development lifecycle is critical for mitigating the long-term negative impacts on software systems. Code review serves as an effective practice that enables developers to check their teammates' code before integration into the codebase. To streamline the generation of review comments, various automated code review approaches have been proposed, where LLM-based methods have significantly advanced the capabilities of automated review generation. However, existing models primarily focus on general-purpose code review, their effectiveness in identifying and addressing security-related issues remains underexplored. Moreover, adapting existing code review approaches to target security issues faces substantial challenges, including data scarcity and inadequate evaluation metrics. To address these limitations, we propose SecureReviewer, a new approach designed for enhancing LLMs' ability to identify and resolve security-related issues during code review. Specifically, we first construct a dataset tailored for training and evaluating secure code review capabilities. Leveraging this dataset, we fine-tune LLMs to generate code review comments that can effectively identify security issues and provide fix suggestions with our proposed secure-aware fine-tuning strategy. To mitigate hallucination in LLMs and enhance the reliability of their outputs, we integrate the RAG technique, which grounds the generated comments in domain-specific security knowledge. Additionally, we introduce SecureBLEU, a new evaluation metric designed to assess the effectiveness of review comments in addressing security issues. Experimental results demonstrate that SecureReviewer outperforms state-of-the-art baselines in both security issue detection accuracy and the overall quality and practical utility of generated review comments.

comment: Accepted by ICSE 2026. Code and data: https://github.com/SIMIAO515/SecureReviewer

☆ Robust Graph Condensation via Classification Complexity Mitigation

Jiayi Luo, Qingyun Sun, Beining Yang, Haonan Yuan, Xingcheng Fu, Yanbiao Ma, Jianxin Li, Philip S. Yu

Graph condensation (GC) has gained significant attention for its ability to synthesize smaller yet informative graphs. However, existing studies often overlook the robustness of GC in scenarios where the original graph is corrupted. In such cases, we observe that the performance of GC deteriorates significantly, while existing robust graph learning technologies offer only limited effectiveness. Through both empirical investigation and theoretical analysis, we reveal that GC is inherently an intrinsic-dimension-reducing process, synthesizing a condensed graph with lower classification complexity. Although this property is critical for effective GC performance, it remains highly vulnerable to adversarial perturbations. To tackle this vulnerability and improve GC robustness, we adopt the geometry perspective of graph data manifold and propose a novel Manifold-constrained Robust Graph Condensation framework named MRGC. Specifically, we introduce three graph data manifold learning modules that guide the condensed graph to lie within a smooth, low-dimensional manifold with minimal class ambiguity, thereby preserving the classification complexity reduction capability of GC and ensuring robust performance under universal adversarial attacks. Extensive experiments demonstrate the robustness of \ModelName\ across diverse attack scenarios.

☆ Personalized Treatment Outcome Prediction from Scarce Data via Dual-Channel Knowledge Distillation and Adaptive Fusion

Wenjie Chen, Li Zhuang, Ziying Luo, Yu Liu, Jiahao Wu, Shengcai Liu

Personalized treatment outcome prediction based on trial data for small-sample and rare patient groups is critical in precision medicine. However, the costly trial data limit the prediction performance. To address this issue, we propose a cross-fidelity knowledge distillation and adaptive fusion network (CFKD-AFN), which leverages abundant but low-fidelity simulation data to enhance predictions on scarce but high-fidelity trial data. CFKD-AFN incorporates a dual-channel knowledge distillation module to extract complementary knowledge from the low-fidelity model, along with an attention-guided fusion module to dynamically integrate multi-source information. Experiments on treatment outcome prediction for the chronic obstructive pulmonary disease demonstrates significant improvements of CFKD-AFN over state-of-the-art methods in prediction accuracy, ranging from 6.67\% to 74.55\%, and strong robustness to varying high-fidelity dataset sizes. Furthermore, we extend CFKD-AFN to an interpretable variant, enabling the exploration of latent medical semantics to support clinical decision-making.

☆ SSCL-BW: Sample-Specific Clean-Label Backdoor Watermarking for Dataset Ownership Verification

Yingjia Wang, Ting Qiao, Xing Liu, Chongzuo Li, Sixing Wu, Jianbin Li

The rapid advancement of deep neural networks (DNNs) heavily relies on large-scale, high-quality datasets. However, unauthorized commercial use of these datasets severely violates the intellectual property rights of dataset owners. Existing backdoor-based dataset ownership verification methods suffer from inherent limitations: poison-label watermarks are easily detectable due to label inconsistencies, while clean-label watermarks face high technical complexity and failure on high-resolution images. Moreover, both approaches employ static watermark patterns that are vulnerable to detection and removal. To address these issues, this paper proposes a sample-specific clean-label backdoor watermarking (i.e., SSCL-BW). By training a U-Net-based watermarked sample generator, this method generates unique watermarks for each sample, fundamentally overcoming the vulnerability of static watermark patterns. The core innovation lies in designing a composite loss function with three components: target sample loss ensures watermark effectiveness, non-target sample loss guarantees trigger reliability, and perceptual similarity loss maintains visual imperceptibility. During ownership verification, black-box testing is employed to check whether suspicious models exhibit predefined backdoor behaviors. Extensive experiments on benchmark datasets demonstrate the effectiveness of the proposed method and its robustness against potential watermark removal attacks.

comment: 8 pages,9 figures

☆ Chain-of-Thought Hijacking

Jianli Zhao, Tingchen Fu, Rylan Schaeffer, Mrinank Sharma, Fazl Barez

Large reasoning models (LRMs) achieve higher task performance by allocating more inference-time compute, and prior works suggest this scaled reasoning may also strengthen safety by improving refusal. Yet we find the opposite: the same reasoning can be used to bypass safeguards. We introduce Chain-of-Thought Hijacking, a jailbreak attack on reasoning models. The attack pads harmful requests with long sequences of harmless puzzle reasoning. Across HarmBench, CoT Hijacking reaches a 99%, 94%, 100%, and 94% attack success rate (ASR) on Gemini 2.5 Pro, GPT o4 mini, Grok 3 mini, and Claude 4 Sonnet, respectively - far exceeding prior jailbreak methods for LRMs. To understand the effectiveness of our attack, we turn to a mechanistic analysis, which shows that mid layers encode the strength of safety checking, while late layers encode the verification outcome. Long benign CoT dilutes both signals by shifting attention away from harmful tokens. Targeted ablations of attention heads identified by this analysis causally decrease refusal, confirming their role in a safety subnetwork. These results show that the most interpretable form of reasoning - explicit CoT - can itself become a jailbreak vector when combined with final-answer cues. We release prompts, outputs, and judge decisions to facilitate replication.

☆ LoCoT2V-Bench: A Benchmark for Long-Form and Complex Text-to-Video Generation

Xiangqing Zheng, Chengyue Wu, Kehai Chen, Min Zhang

Recently text-to-video generation has made impressive progress in producing short, high-quality clips, but evaluating long-form outputs remains a major challenge especially when processing complex prompts. Existing benchmarks mostly rely on simplified prompts and focus on low-level metrics, overlooking fine-grained alignment with prompts and abstract dimensions such as narrative coherence and thematic expression. To address these gaps, we propose LoCoT2V-Bench, a benchmark specifically designed for long video generation (LVG) under complex input conditions. Based on various real-world videos, LoCoT2V-Bench introduces a suite of realistic and complex prompts incorporating elements like scene transitions and event dynamics. Moreover, it constructs a multi-dimensional evaluation framework that includes our newly proposed metrics such as event-level alignment, fine-grained temporal consistency, content clarity, and the Human Expectation Realization Degree (HERD) that focuses on more abstract attributes like narrative flow, emotional response, and character development. Using this framework, we conduct a comprehensive evaluation of nine representative LVG models, finding that while current methods perform well on basic visual and temporal aspects, they struggle with inter-event consistency, fine-grained alignment, and high-level thematic adherence, etc. Overall, LoCoT2V-Bench provides a comprehensive and reliable platform for evaluating long-form complex text-to-video generation and highlights critical directions for future method improvement.

☆ MedSAE: Dissecting MedCLIP Representations with Sparse Autoencoders

Riccardo Renzulli, Colas Lepoutre, Enrico Cassano, Marco Grangetto

Artificial intelligence in healthcare requires models that are accurate and interpretable. We advance mechanistic interpretability in medical vision by applying Medical Sparse Autoencoders (MedSAEs) to the latent space of MedCLIP, a vision-language model trained on chest radiographs and reports. To quantify interpretability, we propose an evaluation framework that combines correlation metrics, entropy analyzes, and automated neuron naming via the MedGEMMA foundation model. Experiments on the CheXpert dataset show that MedSAE neurons achieve higher monosemanticity and interpretability than raw MedCLIP features. Our findings bridge high-performing medical AI and transparency, offering a scalable step toward clinically reliable representations.

☆ Human-in-the-loop Online Rejection Sampling for Robotic Manipulation

Guanxing Lu, Rui Zhao, Haitao Lin, He Zhang, Yansong Tang

Reinforcement learning (RL) is widely used to produce robust robotic manipulation policies, but fine-tuning vision-language-action (VLA) models with RL can be unstable due to inaccurate value estimates and sparse supervision at intermediate steps. In contrast, imitation learning (IL) is easy to train but often underperforms due to its offline nature. In this paper, we propose Hi-ORS, a simple yet effective post-training method that utilizes rejection sampling to achieve both training stability and high robustness. Hi-ORS stabilizes value estimation by filtering out negatively rewarded samples during online fine-tuning, and adopts a reward-weighted supervised training objective to provide dense intermediate-step supervision. For systematic study, we develop an asynchronous inference-training framework that supports flexible online human-in-the-loop corrections, which serve as explicit guidance for learning error-recovery behaviors. Across three real-world tasks and two embodiments, Hi-ORS fine-tunes a pi-base policy to master contact-rich manipulation in just 1.5 hours of real-world training, outperforming RL and IL baselines by a substantial margin in both effectiveness and efficiency. Notably, the fine-tuned policy exhibits strong test-time scalability by reliably executing complex error-recovery behaviors to achieve better performance.

comment: 8 pages

☆ Autograder+: A Multi-Faceted AI Framework for Rich Pedagogical Feedback in Programming Education

Vikrant Sahu, Gagan Raj Gupta, Raghav Borikar, Nitin Mane

The rapid growth of programming education has outpaced traditional assessment tools, leaving faculty with limited means to provide meaningful, scalable feedback. Conventional autograders, while efficient, act as black-box systems that simply return pass/fail results, offering little insight into student thinking or learning needs. Autograder+ is designed to shift autograding from a purely summative process to a formative learning experience. It introduces two key capabilities: automated feedback generation using a fine-tuned Large Language Model, and visualization of student code submissions to uncover learning patterns. The model is fine-tuned on curated student code and expert feedback to ensure pedagogically aligned, context-aware guidance. In evaluation across 600 student submissions from multiple programming tasks, the system produced feedback with strong semantic alignment to instructor comments. For visualization, contrastively learned code embeddings trained on 1,000 annotated submissions enable grouping solutions into meaningful clusters based on functionality and approach. The system also supports prompt-pooling, allowing instructors to guide feedback style through selected prompt templates. By integrating AI-driven feedback, semantic clustering, and interactive visualization, Autograder+ reduces instructor workload while supporting targeted instruction and promoting stronger learning outcomes.

☆ A Pragmatic View of AI Personhood

Joel Z. Leibo, Alexander Sasha Vezhnevets, William A. Cunningham, Stanley M. Bileschi

The emergence of agentic Artificial Intelligence (AI) is set to trigger a "Cambrian explosion" of new kinds of personhood. This paper proposes a pragmatic framework for navigating this diversification by treating personhood not as a metaphysical property to be discovered, but as a flexible bundle of obligations (rights and responsibilities) that societies confer upon entities for a variety of reasons, especially to solve concrete governance problems. We argue that this traditional bundle can be unbundled, creating bespoke solutions for different contexts. This will allow for the creation of practical tools -- such as facilitating AI contracting by creating a target "individual" that can be sanctioned -- without needing to resolve intractable debates about an AI's consciousness or rationality. We explore how individuals fit in to social roles and discuss the use of decentralized digital identity technology, examining both "personhood as a problem", where design choices can create "dark patterns" that exploit human social heuristics, and "personhood as a solution", where conferring a bundle of obligations is necessary to ensure accountability or prevent conflict. By rejecting foundationalist quests for a single, essential definition of personhood, this paper offers a more pragmatic and flexible way to think about integrating AI agents into our society.

comment: 40 pages

☆ SPG-CDENet: Spatial Prior-Guided Cross Dual Encoder Network for Multi-Organ Segmentation

Xizhi Tian, Changjun Zhou, Yulin. Yang

Multi-organ segmentation is a critical task in computer-aided diagnosis. While recent deep learning methods have achieved remarkable success in image segmentation, huge variations in organ size and shape challenge their effectiveness in multi-organ segmentation. To address these challenges, we propose a Spatial Prior-Guided Cross Dual Encoder Network (SPG-CDENet), a novel two-stage segmentation paradigm designed to improve multi-organ segmentation accuracy. Our SPG-CDENet consists of two key components: a spatial prior network and a cross dual encoder network. The prior network generates coarse localization maps that delineate the approximate ROI, serving as spatial guidance for the dual encoder network. The cross dual encoder network comprises four essential components: a global encoder, a local encoder, a symmetric cross-attention module, and a flow-based decoder. The global encoder captures global semantic features from the entire image, while the local encoder focuses on features from the prior network. To enhance the interaction between the global and local encoders, a symmetric cross-attention module is proposed across all layers of the encoders to fuse and refine features. Furthermore, the flow-based decoder directly propagates high-level semantic features from the final encoder layer to all decoder layers, maximizing feature preservation and utilization. Extensive qualitative and quantitative experiments on two public datasets demonstrate the superior performance of SPG-CDENet compared to existing segmentation methods. Furthermore, ablation studies further validate the effectiveness of the proposed modules in improving segmentation accuracy.

☆ Scales++: Compute Efficient Evaluation Subset Selection with Cognitive Scales Embeddings

Andrew M. Bean, Nabeel Seedat, Shengzhuang Chen, Jonathan Richard Schwarz

The prohibitive cost of evaluating large language models (LLMs) on comprehensive benchmarks necessitates the creation of small yet representative data subsets (i.e., tiny benchmarks) that enable efficient assessment while retaining predictive fidelity. Current methods for this task operate under a model-centric paradigm, selecting benchmarking items based on the collective performance of existing models. Such approaches are limited by large upfront costs, an inability to immediately handle new benchmarks (`cold-start'), and the fragile assumption that future models will share the failure patterns of their predecessors. In this work, we challenge this paradigm and propose a item-centric approach to benchmark subset selection, arguing that selection should be based on the intrinsic properties of the task items themselves, rather than on model-specific failure patterns. We instantiate this item-centric efficient benchmarking approach via a novel method, Scales++, where data selection is based on the cognitive demands of the benchmark samples. Empirically, we show Scales++ reduces the upfront selection cost by over 18x while achieving competitive predictive fidelity. On the Open LLM Leaderboard, using just a 0.5\% data subset, we predict full benchmark scores with a 2.9% mean absolute error. We demonstrate that this item-centric approach enables more efficient model evaluation without significant fidelity degradation, while also providing better cold-start performance and more interpretable benchmarking.

comment: 9 pages, 2 figures, 4 tables

☆ AI Mathematician as a Partner in Advancing Mathematical Discovery -- A Case Study in Homogenization Theory

Yuanhang Liu, Beichen Wang, Peng Li, Yang Liu

Artificial intelligence (AI) has demonstrated impressive progress in mathematical reasoning, yet its integration into the practice of mathematical research remains limited. In this study, we investigate how the AI Mathematician (AIM) system can operate as a research partner rather than a mere problem solver. Focusing on a challenging problem in homogenization theory, we analyze the autonomous reasoning trajectories of AIM and incorporate targeted human interventions to structure the discovery process. Through iterative decomposition of the problem into tractable subgoals, selection of appropriate analytical methods, and validation of intermediate results, we reveal how human intuition and machine computation can complement one another. This collaborative paradigm enhances the reliability, transparency, and interpretability of the resulting proofs, while retaining human oversight for formal rigor and correctness. The approach leads to a complete and verifiable proof, and more broadly, demonstrates how systematic human-AI co-reasoning can advance the frontier of mathematical discovery.

comment: 52 pages, 1 figure

☆ BOTS: A Unified Framework for Bayesian Online Task Selection in LLM Reinforcement Finetuning

Qianli Shen, Daoyuan Chen, Yilun Huang, Zhenqing Ling, Yaliang Li, Bolin Ding, Jingren Zhou

Reinforcement finetuning (RFT) is a key technique for aligning Large Language Models (LLMs) with human preferences and enhancing reasoning, yet its effectiveness is highly sensitive to which tasks are explored during training. Uniform task sampling is inefficient, wasting computation on tasks that are either trivial or unsolvable, while existing task selection methods often suffer from high rollout costs, poor adaptivity, or incomplete evidence. We introduce \textbf{BOTS}, a unified framework for \textbf{B}ayesian \textbf{O}nline \textbf{T}ask \textbf{S}election in LLM reinforcement finetuning. Grounded in Bayesian inference, BOTS adaptively maintains posterior estimates of task difficulty as the model evolves. It jointly incorporates \emph{explicit evidence} from direct evaluations of selected tasks and \emph{implicit evidence} inferred from these evaluations for unselected tasks, with Thompson sampling ensuring a principled balance between exploration and exploitation. To make implicit evidence practical, we instantiate it with an ultra-light interpolation-based plug-in that estimates difficulties of unevaluated tasks without extra rollouts, adding negligible overhead. Empirically, across diverse domains and LLM scales, BOTS consistently improves data efficiency and performance over baselines and ablations, providing a practical and extensible solution for dynamic task selection in RFT.

☆ The Geometry of Dialogue: Graphing Language Models to Reveal Synergistic Teams for Multi-Agent Collaboration

Kotaro Furuya, Yuichi Kitagawa

While a multi-agent approach based on large language models (LLMs) represents a promising strategy to surpass the capabilities of single models, its success is critically dependent on synergistic team composition. However, forming optimal teams is a significant challenge, as the inherent opacity of most models obscures the internal characteristics necessary for effective collaboration. In this paper, we propose an interaction-centric framework for automatic team composition that does not require any prior knowledge including their internal architectures, training data, or task performances. Our method constructs a "language model graph" that maps relationships between models from the semantic coherence of pairwise conversations, and then applies community detection to identify synergistic model clusters. Our experiments with diverse LLMs demonstrate that the proposed method discovers functionally coherent groups that reflect their latent specializations. Priming conversations with specific topics identified synergistic teams which outperform random baselines on downstream benchmarks and achieve comparable accuracy to that of manually-curated teams based on known model specializations. Our findings provide a new basis for the automated design of collaborative multi-agent LLM teams.

☆ Reinforcement Learning for Pollution Detection in a Randomized, Sparse and Nonstationary Environment with an Autonomous Underwater Vehicle

Sebastian Zieglmeier, Niklas Erdmann, Narada D. Warakagoda

Reinforcement learning (RL) algorithms are designed to optimize problem-solving by learning actions that maximize rewards, a task that becomes particularly challenging in random and nonstationary environments. Even advanced RL algorithms are often limited in their ability to solve problems in these conditions. In applications such as searching for underwater pollution clouds with autonomous underwater vehicles (AUVs), RL algorithms must navigate reward-sparse environments, where actions frequently result in a zero reward. This paper aims to address these challenges by revisiting and modifying classical RL approaches to efficiently operate in sparse, randomized, and nonstationary environments. We systematically study a large number of modifications, including hierarchical algorithm changes, multigoal learning, and the integration of a location memory as an external output filter to prevent state revisits. Our results demonstrate that a modified Monte Carlo-based approach significantly outperforms traditional Q-learning and two exhaustive search patterns, illustrating its potential in adapting RL to complex environments. These findings suggest that reinforcement learning approaches can be effectively adapted for use in random, nonstationary, and reward-sparse environments.

☆ Discovering State Equivalences in UCT Search Trees By Action Pruning

Robin Schmöcker, Alexander Dockhorn, Bodo Rosenhahn

One approach to enhance Monte Carlo Tree Search (MCTS) is to improve its sample efficiency by grouping/abstracting states or state-action pairs and sharing statistics within a group. Though state-action pair abstractions are mostly easy to find in algorithms such as On the Go Abstractions in Upper Confidence bounds applied to Trees (OGA-UCT), nearly no state abstractions are found in either noisy or large action space settings due to constraining conditions. We provide theoretical and empirical evidence for this claim, and we slightly alleviate this state abstraction problem by proposing a weaker state abstraction condition that trades a minor loss in accuracy for finding many more abstractions. We name this technique Ideal Pruning Abstractions in UCT (IPA-UCT), which outperforms OGA-UCT (and any of its derivatives) across a large range of test domains and iteration budgets as experimentally validated. IPA-UCT uses a different abstraction framework from Abstraction of State-Action Pairs (ASAP) which is the one used by OGA-UCT, which we name IPA. Furthermore, we show that both IPA and ASAP are special cases of a more general framework that we call p-ASAP which itself is a special case of the ASASAP framework.

☆ MisSynth: Improving MISSCI Logical Fallacies Classification with Synthetic Data

Mykhailo Poliakov, Nadiya Shvai

Health-related misinformation is very prevalent and potentially harmful. It is difficult to identify, especially when claims distort or misinterpret scientific findings. We investigate the impact of synthetic data generation and lightweight fine-tuning techniques on the ability of large language models (LLMs) to recognize fallacious arguments using the MISSCI dataset and framework. In this work, we propose MisSynth, a pipeline that applies retrieval-augmented generation (RAG) to produce synthetic fallacy samples, which are then used to fine-tune an LLM model. Our results show substantial accuracy gains with fine-tuned models compared to vanilla baselines. For instance, the LLaMA 3.1 8B fine-tuned model achieved an over 35% F1-score absolute improvement on the MISSCI test split over its vanilla baseline. We demonstrate that introducing synthetic fallacy data to augment limited annotated resources can significantly enhance zero-shot LLM classification performance on real-world scientific misinformation tasks, even with limited computational resources. The code and synthetic dataset are available on https://github.com/mxpoliakov/MisSynth.

☆ Linear Causal Discovery with Interventional Constraints

Zhigao Guo, Feng Dong

Incorporating causal knowledge and mechanisms is essential for refining causal models and improving downstream tasks such as designing new treatments. In this paper, we introduce a novel concept in causal discovery, termed interventional constraints, which differs fundamentally from interventional data. While interventional data require direct perturbations of variables, interventional constraints encode high-level causal knowledge in the form of inequality constraints on causal effects. For instance, in the Sachs dataset (Sachs et al.\ 2005), Akt has been shown to be activated by PIP3, meaning PIP3 exerts a positive causal effect on Akt. Existing causal discovery methods allow enforcing structural constraints (for example, requiring a causal path from PIP3 to Akt), but they may still produce incorrect causal conclusions such as learning that "PIP3 inhibits Akt". Interventional constraints bridge this gap by explicitly constraining the total causal effect between variable pairs, ensuring learned models respect known causal influences. To formalize interventional constraints, we propose a metric to quantify total causal effects for linear causal models and formulate the problem as a constrained optimization task, solved using a two-stage constrained optimization method. We evaluate our approach on real-world datasets and demonstrate that integrating interventional constraints not only improves model accuracy and ensures consistency with established findings, making models more explainable, but also facilitates the discovery of new causal relationships that would otherwise be costly to identify.

☆ GLYPH-SR: Can We Achieve Both High-Quality Image Super-Resolution and High-Fidelity Text Recovery via VLM-guided Latent Diffusion Model? ICLR 2026

Mingyu Sung, Seungjae Ham, Kangwoo Kim, Yeokyoung Yoon, Sangseok Yun, Il-Min Kim, Jae-Mo Kang

Image super-resolution(SR) is fundamental to many vision system-from surveillance and autonomy to document analysis and retail analytics-because recovering high-frequency details, especially scene-text, enables reliable downstream perception. Scene-text, i.e., text embedded in natural images such as signs, product labels, and storefronts, often carries the most actionable information; when characters are blurred or hallucinated, optical character recognition(OCR) and subsequent decisions fail even if the rest of the image appears sharp. Yet previous SR research has often been tuned to distortion (PSNR/SSIM) or learned perceptual metrics (LIPIS, MANIQA, CLIP-IQA, MUSIQ) that are largely insensitive to character-level errors. Furthermore, studies that do address text SR often focus on simplified benchmarks with isolated characters, overlooking the challenges of text within complex natural scenes. As a result, scene-text is effectively treated as generic texture. For SR to be effective in practical deployments, it is therefore essential to explicitly optimize for both text legibility and perceptual quality. We present GLYPH-SR, a vision-language-guided diffusion framework that aims to achieve both objectives jointly. GLYPH-SR utilizes a Text-SR Fusion ControlNet(TS-ControlNet) guided by OCR data, and a ping-pong scheduler that alternates between text- and scene-centric guidance. To enable targeted text restoration, we train these components on a synthetic corpus while keeping the main SR branch frozen. Across SVT, SCUT-CTW1500, and CUTE80 at x4, and x8, GLYPH-SR improves OCR F1 by up to +15.18 percentage points over diffusion/GAN baseline (SVT x8, OpenOCR) while maintaining competitive MANIQA, CLIP-IQA, and MUSIQ. GLYPH-SR is designed to satisfy both objectives simultaneously-high readability and high visual realism-delivering SR that looks right and reds right.

comment: 11 pages, 6 figures. Includes supplementary material. Under review as a conference paper at ICLR 2026

☆ From Amateur to Master: Infusing Knowledge into LLMs via Automated Curriculum Learning

Nishit Neema, Srinjoy Mukherjee, Sapan Shah, Gokul Ramakrishnan, Ganesh Venkatesh

Large Language Models (LLMs) excel at general tasks but underperform in specialized domains like economics and psychology, which require deep, principled understanding. To address this, we introduce ACER (Automated Curriculum-Enhanced Regimen) that transforms generalist models into domain experts without sacrificing their broad capabilities. ACER first synthesizes a comprehensive, textbook-style curriculum by generating a table of contents for a subject and then creating question-answer (QA) pairs guided by Bloom's taxonomy. This ensures systematic topic coverage and progressively increasing difficulty. The resulting synthetic corpus is used for continual pretraining with an interleaved curriculum schedule, aligning learning across both content and cognitive dimensions. Experiments with Llama 3.2 (1B and 3B) show significant gains in specialized MMLU subsets. In challenging domains like microeconomics, where baselines struggle, ACER boosts accuracy by 5 percentage points. Across all target domains, we observe a consistent macro-average improvement of 3 percentage points. Notably, ACER not only prevents catastrophic forgetting but also facilitates positive cross-domain knowledge transfer, improving performance on non-target domains by 0.7 points. Beyond MMLU, ACER enhances performance on knowledge-intensive benchmarks like ARC and GPQA by over 2 absolute points, while maintaining stable performance on general reasoning tasks. Our results demonstrate that ACER offers a scalable and effective recipe for closing critical domain gaps in LLMs.

☆ Posterior Sampling by Combining Diffusion Models with Annealed Langevin Dynamics NeurIPS 2025

Zhiyang Xun, Shivam Gupta, Eric Price

Given a noisy linear measurement $y = Ax + \xi$ of a distribution $p(x)$, and a good approximation to the prior $p(x)$, when can we sample from the posterior $p(x \mid y)$? Posterior sampling provides an accurate and fair framework for tasks such as inpainting, deblurring, and MRI reconstruction, and several heuristics attempt to approximate it. Unfortunately, approximate posterior sampling is computationally intractable in general. To sidestep this hardness, we focus on (local or global) log-concave distributions $p(x)$. In this regime, Langevin dynamics yields posterior samples when the exact scores of $p(x)$ are available, but it is brittle to score--estimation error, requiring an MGF bound (sub-exponential error). By contrast, in the unconditional setting, diffusion models succeed with only an $L^2$ bound on the score error. We prove that combining diffusion models with an annealed variant of Langevin dynamics achieves conditional sampling in polynomial time using merely an $L^4$ bound on the score error.

comment: NeurIPS 2025

☆ GraphCompliance: Aligning Policy and Context Graphs for LLM-Based Regulatory Compliance

Jiseong Chung, Ronny Ko, Wonchul Yoo, Makoto Onizuka, Sungmok Kim, Tae-Wan Kim, Won-Yong Shin

Compliance at web scale poses practical challenges: each request may require a regulatory assessment. Regulatory texts (e.g., the General Data Protection Regulation, GDPR) are cross-referential and normative, while runtime contexts are expressed in unstructured natural language. This setting motivates us to align semantic information in unstructured text with the structured, normative elements of regulations. To this end, we introduce GraphCompliance, a framework that represents regulatory texts as a Policy Graph and runtime contexts as a Context Graph, and aligns them. In this formulation, the policy graph encodes normative structure and cross-references, whereas the context graph formalizes events as subject-action-object (SAO) and entity-relation triples. This alignment anchors the reasoning of a judge large language model (LLM) in structured information and helps reduce the burden of regulatory interpretation and event parsing, enabling a focus on the core reasoning step. In experiments on 300 GDPR-derived real-world scenarios spanning five evaluation tasks, GraphCompliance yields 4.1-7.2 percentage points (pp) higher micro-F1 than LLM-only and RAG baselines, with fewer under- and over-predictions, resulting in higher recall and lower false positive rates. Ablation studies indicate contributions from each graph component, suggesting that structured representations and a judge LLM are complementary for normative reasoning.

comment: Under review at The Web Conference 2026 (Semantics & Knowledge track). Code will be released upon acceptance. This arXiv v1 contains no repository links to preserve double-blind review

☆ Implicit Bias of Per-sample Adam on Separable Data: Departure from the Full-batch Regime

Beomhan Baek, Minhak Song, Chulhee Yun

Adam [Kingma and Ba, 2015] is the de facto optimizer in deep learning, yet its theoretical understanding remains limited. Prior analyses show that Adam favors solutions aligned with $\ell_\infty$-geometry, but these results are restricted to the full-batch regime. In this work, we study the implicit bias of incremental Adam (using one sample per step) for logistic regression on linearly separable data, and we show that its bias can deviate from the full-batch behavior. To illustrate this, we construct a class of structured datasets where incremental Adam provably converges to the $\ell_2$-max-margin classifier, in contrast to the $\ell_\infty$-max-margin bias of full-batch Adam. For general datasets, we develop a proxy algorithm that captures the limiting behavior of incremental Adam as $\beta_2 \to 1$ and we characterize its convergence direction via a data-dependent dual fixed-point formulation. Finally, we prove that, unlike Adam, Signum [Bernstein et al., 2018] converges to the $\ell_\infty$-max-margin classifier for any batch size by taking $\beta$ close enough to 1. Overall, our results highlight that the implicit bias of Adam crucially depends on both the batching scheme and the dataset, while Signum remains invariant.

comment: 50 pages

☆ Understanding Hardness of Vision-Language Compositionality from A Token-level Causal Lens

Ziliang Chen, Tianang Xiao, Jusheng Zhang, Yongsen Zheng, Xipeng Chen

Contrastive Language-Image Pre-training (CLIP) delivers strong cross modal generalization by aligning images and texts in a shared embedding space, yet it persistently fails at compositional reasoning over objects, attributes, and relations often behaving like a bag-of-words matcher. Prior causal accounts typically model text as a single vector, obscuring token-level structure and leaving core phenomena-such as prompt sensitivity and failures on hard negatives unexplained. We address this gap with a token-aware causal representation learning (CRL) framework grounded in a sequential, language-token SCM. Our theory extends block identifiability to tokenized text, proving that CLIP's contrastive objective can recover the modal-invariant latent variable under both sentence-level and token-level SCMs. Crucially, token granularity yields the first principled explanation of CLIP's compositional brittleness: composition nonidentifiability. We show the existence of pseudo-optimal text encoders that achieve perfect modal-invariant alignment yet are provably insensitive to SWAP, REPLACE, and ADD operations over atomic concepts, thereby failing to distinguish correct captions from hard negatives despite optimizing the same training objective as true-optimal encoders. The analysis further links language-side nonidentifiability to visual-side failures via the modality gap and shows how iterated composition operators compound hardness, motivating improved negative mining strategies.

☆ Can Agent Conquer Web? Exploring the Frontiers of ChatGPT Atlas Agent in Web Games

Jingran Zhang, Ning Li, Justin Cui

OpenAI's ChatGPT Atlas introduces new capabilities for web interaction, enabling the model to analyze webpages, process user intents, and execute cursor and keyboard inputs directly within the browser. While its capacity for information retrieval tasks has been demonstrated, its performance in dynamic, interactive environments remains less explored. In this study, we conduct an early evaluation of Atlas's web interaction capabilities using browser-based games as test scenarios, including Google's T-Rex Runner, Sudoku, Flappy Bird, and Stein.world. We employ in-game performance scores as quantitative metrics to assess performance across different task types. Our results show that Atlas performs strongly in logical reasoning tasks like Sudoku, completing puzzles significantly faster than human baselines, but struggles substantially in real-time games requiring precise timing and motor control, often failing to progress beyond initial obstacles. These findings suggest that while Atlas demonstrates capable analytical processing, there remain notable limitations in dynamic web environments requiring real-time interaction. The website of our project can be found at https://atlas-game-eval.github.io.

☆ Unravelling the Mechanisms of Manipulating Numbers in Language Models

Michal Štefánik, Timothee Mickus, Marek Kadlčík, Bertram Højer, Michal Spiegel, Raúl Vázquez, Aman Sinha, Josef Kuchař, Philipp Mondorf

Recent work has shown that different large language models (LLMs) converge to similar and accurate input embedding representations for numbers. These findings conflict with the documented propensity of LLMs to produce erroneous outputs when dealing with numeric information. In this work, we aim to explain this conflict by exploring how language models manipulate numbers and quantify the lower bounds of accuracy of these mechanisms. We find that despite surfacing errors, different language models learn interchangeable representations of numbers that are systematic, highly accurate and universal across their hidden states and the types of input contexts. This allows us to create universal probes for each LLM and to trace information -- including the causes of output errors -- to specific layers. Our results lay a fundamental understanding of how pre-trained LLMs manipulate numbers and outline the potential of more accurate probing techniques in addressed refinements of LLMs' architectures.

☆ Distributional Multi-objective Black-box Optimization for Diffusion-model Inference-time Multi-Target Generation

Kim Yong Tan, Yueming Lyu, Ivor Tsang, Yew-Soon Ong

Diffusion models have been successful in learning complex data distributions. This capability has driven their application to high-dimensional multi-objective black-box optimization problem. Existing approaches often employ an external optimization loop, such as an evolutionary algorithm, to the diffusion model. However, these approaches treat the diffusion model as a black-box refiner, which overlooks the internal distribution transition of the diffusion generation process, limiting their efficiency. To address these challenges, we propose the Inference-time Multi-target Generation (IMG) algorithm, which optimizes the diffusion process at inference-time to generate samples that simultaneously satisfy multiple objectives. Specifically, our IMG performs weighted resampling during the diffusion generation process according to the expected aggregated multi-objective values. This weighted resampling strategy ensures the diffusion-generated samples are distributed according to our desired multi-target Boltzmann distribution. We further derive that the multi-target Boltzmann distribution has an interesting log-likelihood interpretation, where it is the optimal solution to the distributional multi-objective optimization problem. We implemented IMG for a multi-objective molecule generation task. Experiments show that IMG, requiring only a single generation pass, achieves a significantly higher hypervolume than baseline optimization algorithms that often require hundreds of diffusion generations. Notably, our algorithm can be viewed as an optimized diffusion process and can be integrated into existing methods to further improve their performance.

☆ A Research Roadmap for Augmenting Software Engineering Processes and Software Products with Generative AI

Domenico Amalfitano, Andreas Metzger, Marco Autili, Tommaso Fulcini, Tobias Hey, Jan Keim, Patrizio Pelliccione, Vincenzo Scotti, Anne Koziolek, Raffaela Mirandola, Andreas Vogelsang

Generative AI (GenAI) is rapidly transforming software engineering (SE) practices, influencing how SE processes are executed, as well as how software systems are developed, operated, and evolved. This paper applies design science research to build a roadmap for GenAI-augmented SE. The process consists of three cycles that incrementally integrate multiple sources of evidence, including collaborative discussions from the FSE 2025 "Software Engineering 2030" workshop, rapid literature reviews, and external feedback sessions involving peers. McLuhan's tetrads were used as a conceptual instrument to systematically capture the transforming effects of GenAI on SE processes and software products.The resulting roadmap identifies four fundamental forms of GenAI augmentation in SE and systematically characterizes their related research challenges and opportunities. These insights are then consolidated into a set of future research directions. By grounding the roadmap in a rigorous multi-cycle process and cross-validating it among independent author teams and peers, the study provides a transparent and reproducible foundation for analyzing how GenAI affects SE processes, methods and tools, and for framing future research within this rapidly evolving area. Based on these findings, the article finally makes ten predictions for SE in the year 2030.

☆ Graph-Enhanced Policy Optimization in LLM Agent Training

Jiazhen Yuan, Wei Zhao, Zhengbiao Bai

Group based reinforcement learning (RL) has shown impressive results on complex reasoning and mathematical tasks. Yet, when applied to train multi-turn, interactive LLM agents, these methods often suffer from structural blindness-the inability to exploit the underlying connectivity of the environment. This manifests in three critical challenges: (1) inefficient, unguided exploration, (2) imprecise credit assignment due to overlooking pivotal states, and (3) myopic planning caused by static reward discounting. We address these issues with Graph-Enhanced Policy Optimization (GEPO), which dynamically constructs a state-transition graph from agent experience and employs graph-theoretic centrality to provide three synergistic learning signals: (1)structured intrinsic rewards that guide exploration toward high-impact states, (2) a graph-enhanced advantage function for topology-aware credit assignment, and (3) a dynamic discount factor adapted to each state's strategic value. On the ALFWorld, WebShop, and a proprietary Workbench benchmarks, GEPO demonstrates strong performance, achieving absolute success rate gains of +4.1%, +5.3%, and +10.9% over competitive baselines. These results highlight that explicitly modeling environmental structure is a robust, generalizable strategy for advancing LLM agent training.

comment: Under review as a conference paper

☆ Angular Steering: Behavior Control via Rotation in Activation Space NeurIPS 2025

Hieu M. Vu, Tan M. Nguyen

Controlling specific behaviors in large language models while preserving their general capabilities is a central challenge for safe and reliable artificial intelligence deployment. Current steering methods, such as vector addition and directional ablation, are constrained within a two-dimensional subspace defined by the activation and feature direction, making them sensitive to chosen parameters and potentially affecting unrelated features due to unintended interactions in activation space. We introduce Angular Steering, a novel and flexible method for behavior modulation that operates by rotating activations within a fixed two-dimensional subspace. By formulating steering as a geometric rotation toward or away from a target behavior direction, Angular Steering provides continuous, fine-grained control over behaviors such as refusal and compliance. We demonstrate this method using refusal steering emotion steering as use cases. Additionally, we propose Adaptive Angular Steering, a selective variant that rotates only activations aligned with the target feature, further enhancing stability and coherence. Angular Steering generalizes existing addition and orthogonalization techniques under a unified geometric rotation framework, simplifying parameter selection and maintaining model stability across a broader range of adjustments. Experiments across multiple model families and sizes show that Angular Steering achieves robust behavioral control while maintaining general language modeling performance, underscoring its flexibility, generalization, and robustness compared to prior approaches. Code and artifacts are available at https://github.com/lone17/angular-steering/.

comment: NeurIPS 2025 (Spotlight)

☆ Retrieval Augmented Generation-Enhanced Distributed LLM Agents for Generalizable Traffic Signal Control with Emergency Vehicles

Xinhang Li, Qing Guo, Junyu Chen, Zheng Guo, Shengzhe Xu, Lei Li, Lin Zhang

With increasing urban traffic complexity, Traffic Signal Control (TSC) is essential for optimizing traffic flow and improving road safety. Large Language Models (LLMs) emerge as promising approaches for TSC. However, they are prone to hallucinations in emergencies, leading to unreliable decisions that may cause substantial delays for emergency vehicles. Moreover, diverse intersection types present substantial challenges for traffic state encoding and cross-intersection training, limiting generalization across heterogeneous intersections. Therefore, this paper proposes Retrieval Augmented Generation (RAG)-enhanced distributed LLM agents with Emergency response for Generalizable TSC (REG-TSC). Firstly, this paper presents an emergency-aware reasoning framework, which dynamically adjusts reasoning depth based on the emergency scenario and is equipped with a novel Reviewer-based Emergency RAG (RERAG) to distill specific knowledge and guidance from historical cases, enhancing the reliability and rationality of agents' emergency decisions. Secondly, this paper designs a type-agnostic traffic representation and proposes a Reward-guided Reinforced Refinement (R3) for heterogeneous intersections. R3 adaptively samples training experience from diverse intersections with environment feedback-based priority and fine-tunes LLM agents with a designed reward-weighted likelihood loss, guiding REG-TSC toward high-reward policies across heterogeneous intersections. On three real-world road networks with 17 to 177 heterogeneous intersections, extensive experiments show that REG-TSC reduces travel time by 42.00%, queue length by 62.31%, and emergency vehicle waiting time by 83.16%, outperforming other state-of-the-art methods.

☆ Questionnaire meets LLM: A Benchmark and Empirical Study of Structural Skills for Understanding Questions and Responses

Duc-Hai Nguyen, Vijayakumar Nanjappan, Barry O'Sullivan, Hoang D. Nguyen

Millions of people take surveys every day, from market polls and academic studies to medical questionnaires and customer feedback forms. These datasets capture valuable insights, but their scale and structure present a unique challenge for large language models (LLMs), which otherwise excel at few-shot reasoning over open-ended text. Yet, their ability to process questionnaire data or lists of questions crossed with hundreds of respondent rows remains underexplored. Current retrieval and survey analysis tools (e.g., Qualtrics, SPSS, REDCap) are typically designed for humans in the workflow, limiting such data integration with LLM and AI-empowered automation. This gap leaves scientists, surveyors, and everyday users without evidence-based guidance on how to best represent questionnaires for LLM consumption. We address this by introducing QASU (Questionnaire Analysis and Structural Understanding), a benchmark that probes six structural skills, including answer lookup, respondent count, and multi-hop inference, across six serialization formats and multiple prompt strategies. Experiments on contemporary LLMs show that choosing an effective format and prompt combination can improve accuracy by up to 8.8% points compared to suboptimal formats. For specific tasks, carefully adding a lightweight structural hint through self-augmented prompting can yield further improvements of 3-4% points on average. By systematically isolating format and prompting effects, our open source benchmark offers a simple yet versatile foundation for advancing both research and real-world practice in LLM-based questionnaire analysis.

comment: 14 pages, 3 figures, 8 tables

☆ MPRU: Modular Projection-Redistribution Unlearning as Output Filter for Classification Pipelines

Minyi Peng, Darian Gunamardi, Ivan Tjuawinata, Kwok-Yan Lam

As a new and promising approach, existing machine unlearning (MU) works typically emphasize theoretical formulations or optimization objectives to achieve knowledge removal. However, when deployed in real-world scenarios, such solutions typically face scalability issues and have to address practical requirements such as full access to original datasets and model. In contrast to the existing approaches, we regard classification training as a sequential process where classes are learned sequentially, which we call \emph{inductive approach}. Unlearning can then be done by reversing the last training sequence. This is implemented by appending a projection-redistribution layer in the end of the model. Such an approach does not require full access to the original dataset or the model, addressing the challenges of existing methods. This enables modular and model-agnostic deployment as an output filter into existing classification pipelines with minimal alterations. We conducted multiple experiments across multiple datasets including image (CIFAR-10/100 using CNN-based model) and tabular datasets (Covertype using tree-based model). Experiment results show consistently similar output to a fully retrained model with a high computational cost reduction. This demonstrates the applicability, scalability, and system compatibility of our solution while maintaining the performance of the output in a more practical setting.

comment: 10 pages, 6 figures

☆ Test-Time Alignment of LLMs via Sampling-Based Optimal Control in pre-logit space

Sekitoshi Kanai, Tsukasa Yoshida, Hiroshi Takahashi, Haru Kuroki, Kazumune Hashimoto

Test-time alignment of large language models (LLMs) attracts attention because fine-tuning LLMs requires high computational costs. In this paper, we propose a new test-time alignment method called adaptive importance sampling on pre-logits (AISP) on the basis of the sampling-based model predictive control with the stochastic control input. AISP applies the Gaussian perturbation into pre-logits, which are outputs of the penultimate layer, so as to maximize expected rewards with respect to the mean of the perturbation. We demonstrate that the optimal mean is obtained by importance sampling with sampled rewards. AISP outperforms best-of-n sampling in terms of rewards over the number of used samples and achieves higher rewards than other reward-based test-time alignment methods.

comment: 21 pages, 8 figures

☆ Hybrid LLM and Higher-Order Quantum Approximate Optimization for CSA Collateral Management

Tao Jin, Stuart Florescu, Heyu, Jin

We address finance-native collateral optimization under ISDA Credit Support Annexes (CSAs), where integer lots, Schedule A haircuts, RA/MTA gating, and issuer/currency/class caps create rugged, legally bounded search spaces. We introduce a certifiable hybrid pipeline purpose-built for this domain: (i) an evidence-gated LLM that extracts CSA terms to a normalized JSON (abstain-by-default, span-cited); (ii) a quantum-inspired explorer that interleaves simulated annealing with micro higher order QAOA (HO-QAOA) on binding sub-QUBOs (subset size n <= 16, order k <= 4) to coordinate multi-asset moves across caps and RA-induced discreteness; (iii) a weighted risk-aware objective (Movement, CVaR, funding-priced overshoot) with an explicit coverage window U <= Reff+B; and (iv) CP-SAT as single arbiter to certify feasibility and gaps, including a U-cap pre-check that reports the minimal feasible buffer B*. Encoding caps/rounding as higher-order terms lets HO-QAOA target the domain couplings that defeat local swaps. On government bond datasets and multi-CSA inputs, the hybrid improves a strong classical baseline (BL-3) by 9.1%, 9.6%, and 10.7% across representative harnesses, delivering better cost-movement-tail frontiers under governance settings. We release governance grade artifacts-span citations, valuation matrix audit, weight provenance, QUBO manifests, and CP-SAT traces-to make results auditable and reproducible.

comment: 6 pages

☆ Towards Global Retrieval Augmented Generation: A Benchmark for Corpus-Level Reasoning

Qi Luo, Xiaonan Li, Tingshuo Fan, Xinchi Chen, Xipeng Qiu

Retrieval-augmented generation (RAG) has emerged as a leading approach to reducing hallucinations in large language models (LLMs). Current RAG evaluation benchmarks primarily focus on what we call local RAG: retrieving relevant chunks from a small subset of documents to answer queries that require only localized understanding within specific text chunks. However, many real-world applications require a fundamentally different capability -- global RAG -- which involves aggregating and analyzing information across entire document collections to derive corpus-level insights (for example, "What are the top 10 most cited papers in 2023?"). In this paper, we introduce GlobalQA -- the first benchmark specifically designed to evaluate global RAG capabilities, covering four core task types: counting, extremum queries, sorting, and top-k extraction. Through systematic evaluation across different models and baselines, we find that existing RAG methods perform poorly on global tasks, with the strongest baseline achieving only 1.51 F1 score. To address these challenges, we propose GlobalRAG, a multi-tool collaborative framework that preserves structural coherence through chunk-level retrieval, incorporates LLM-driven intelligent filters to eliminate noisy documents, and integrates aggregation modules for precise symbolic computation. On the Qwen2.5-14B model, GlobalRAG achieves 6.63 F1 compared to the strongest baseline's 1.51 F1, validating the effectiveness of our method.

☆ What's In My Human Feedback? Learning Interpretable Descriptions of Preference Data

Rajiv Movva, Smitha Milli, Sewon Min, Emma Pierson

Human feedback can alter language models in unpredictable and undesirable ways, as practitioners lack a clear understanding of what feedback data encodes. While prior work studies preferences over certain attributes (e.g., length or sycophancy), automatically extracting relevant features without pre-specifying hypotheses remains challenging. We introduce What's In My Human Feedback? (WIMHF), a method to explain feedback data using sparse autoencoders. WIMHF characterizes both (1) the preferences a dataset is capable of measuring and (2) the preferences that the annotators actually express. Across 7 datasets, WIMHF identifies a small number of human-interpretable features that account for the majority of the preference prediction signal achieved by black-box models. These features reveal a wide diversity in what humans prefer, and the role of dataset-level context: for example, users on Reddit prefer informality and jokes, while annotators in HH-RLHF and PRISM disprefer them. WIMHF also surfaces potentially unsafe preferences, such as that LMArena users tend to vote against refusals, often in favor of toxic content. The learned features enable effective data curation: re-labeling the harmful examples in Arena yields large safety gains (+37%) with no cost to general performance. They also allow fine-grained personalization: on the Community Alignment dataset, we learn annotator-specific weights over subjective features that improve preference prediction. WIMHF provides a human-centered analysis method for practitioners to better understand and use preference data.

comment: Code: https://github.com/rmovva/wimhf

☆ Don't Let It Fade: Preserving Edits in Diffusion Language Models via Token Timestep Allocation NeurIPS 2025

Woojin Kim, Jaeyoung Do

While diffusion language models (DLMs) enable fine-grained refinement, their practical controllability remains fragile. We identify and formally characterize a central failure mode called update forgetting, in which uniform and context agnostic updates induce token level fluctuations across timesteps, erasing earlier semantic edits and disrupting the cumulative refinement process, thereby degrading fluency and coherence. As this failure originates in uniform and context agnostic updates, effective control demands explicit token ordering. We propose Token Timestep Allocation (TTA), which realizes soft and semantic token ordering via per token timestep schedules: critical tokens are frozen early, while uncertain tokens receive continued refinement. This timestep based ordering can be instantiated as either a fixed policy or an adaptive policy driven by task signals, thereby supporting a broad spectrum of refinement strategies. Because it operates purely at inference time, it applies uniformly across various DLMs and naturally extends to diverse supervision sources. Empirically, TTA improves controllability and fluency: on sentiment control, it yields more than 20 percent higher accuracy and nearly halves perplexity using less than one fifth the steps; in detoxification, it lowers maximum toxicity (12.2 versus 14.5) and perplexity (26.0 versus 32.0). Together, these results demonstrate that softened ordering via timestep allocation is the critical lever for mitigating update forgetting and achieving stable and controllable diffusion text generation.

comment: Accepted in NeurIPS 2025

☆ Predicting All-Cause Hospital Readmissions from Medical Claims Data of Hospitalised Patients

Avinash Kadimisetty, Arun Rajagopalan, Vijendra SK

Reducing preventable hospital readmissions is a national priority for payers, providers, and policymakers seeking to improve health care and lower costs. The rate of readmission is being used as a benchmark to determine the quality of healthcare provided by the hospitals. In thisproject, we have used machine learning techniques like Logistic Regression, Random Forest and Support Vector Machines to analyze the health claims data and identify demographic and medical factors that play a crucial role in predicting all-cause readmissions. As the health claims data is high dimensional, we have used Principal Component Analysis as a dimension reduction technique and used the results for building regression models. We compared and evaluated these models based on the Area Under Curve (AUC) metric. Random Forest model gave the highest performance followed by Logistic Regression and Support Vector Machine models. These models can be used to identify the crucial factors causing readmissions and help identify patients to focus on to reduce the chances of readmission, ultimately bringing down the cost and increasing the quality of healthcare provided to the patients.

comment: NCMLAI 2018

☆ ConceptScope: Characterizing Dataset Bias via Disentangled Visual Concepts NeurIPS 2025

Jinho Choi, Hyesu Lim, Steffen Schneider, Jaegul Choo

Dataset bias, where data points are skewed to certain concepts, is ubiquitous in machine learning datasets. Yet, systematically identifying these biases is challenging without costly, fine-grained attribute annotations. We present ConceptScope, a scalable and automated framework for analyzing visual datasets by discovering and quantifying human-interpretable concepts using Sparse Autoencoders trained on representations from vision foundation models. ConceptScope categorizes concepts into target, context, and bias types based on their semantic relevance and statistical correlation to class labels, enabling class-level dataset characterization, bias identification, and robustness evaluation through concept-based subgrouping. We validate that ConceptScope captures a wide range of visual concepts, including objects, textures, backgrounds, facial attributes, emotions, and actions, through comparisons with annotated datasets. Furthermore, we show that concept activations produce spatial attributions that align with semantically meaningful image regions. ConceptScope reliably detects known biases (e.g., background bias in Waterbirds) and uncovers previously unannotated ones (e.g, co-occurring objects in ImageNet), offering a practical tool for dataset auditing and model diagnostics.

comment: Published in the Thirty-Ninth Conference on Neural Information Processing Systems (NeurIPS 2025)

☆ Accumulative SGD Influence Estimation for Data Attribution

Yunxiao Shi, Shuo Yang, Yixin Su, Rui Zhang, Min Xu

Modern data-centric AI needs precise per-sample influence. Standard SGD-IE approximates leave-one-out effects by summing per-epoch surrogates and ignores cross-epoch compounding, which misranks critical examples. We propose ACC-SGD-IE, a trajectory-aware estimator that propagates the leave-one-out perturbation across training and updates an accumulative influence state at each step. In smooth strongly convex settings it achieves geometric error contraction and, in smooth non-convex regimes, it tightens error bounds; larger mini-batches further reduce constants. Empirically, on Adult, 20 Newsgroups, and MNIST under clean and corrupted data and both convex and non-convex training, ACC-SGD-IE yields more accurate influence estimates, especially over long epochs. For downstream data cleansing it more reliably flags noisy samples, producing models trained on ACC-SGD-IE cleaned data that outperform those cleaned with SGD-IE.

☆ Linking Heterogeneous Data with Coordinated Agent Flows for Social Media Analysis

Shifu Chen, Dazhen Deng, Zhihong Xu, Sijia Xu, Tai-Quan Peng, Yingcai Wu

Social media platforms generate massive volumes of heterogeneous data, capturing user behaviors, textual content, temporal dynamics, and network structures. Analyzing such data is crucial for understanding phenomena such as opinion dynamics, community formation, and information diffusion. However, discovering insights from this complex landscape is exploratory, conceptually challenging, and requires expertise in social media mining and visualization. Existing automated approaches, though increasingly leveraging large language models (LLMs), remain largely confined to structured tabular data and cannot adequately address the heterogeneity of social media analysis. We present SIA (Social Insight Agents), an LLM agent system that links heterogeneous multi-modal data -- including raw inputs (e.g., text, network, and behavioral data), intermediate outputs, mined analytical results, and visualization artifacts -- through coordinated agent flows. Guided by a bottom-up taxonomy that connects insight types with suitable mining and visualization techniques, SIA enables agents to plan and execute coherent analysis strategies. To ensure multi-modal integration, it incorporates a data coordinator that unifies tabular, textual, and network data into a consistent flow. Its interactive interface provides a transparent workflow where users can trace, validate, and refine the agent's reasoning, supporting both adaptability and trustworthiness. Through expert-centered case studies and quantitative evaluation, we show that SIA effectively discovers diverse and meaningful insights from social media while supporting human-agent collaboration in complex analytical tasks.

☆ One Model to Critique Them All: Rewarding Agentic Tool-Use via Efficient Reasoning

Renhao Li, Jianhong Tu, Yang Su, Hamid Alinejad-Rokny, Derek F. Wong, Junyang Lin, Min Yang

Reward models (RMs) play a critical role in aligning large language models (LLMs) with human preferences. Yet in the domain of tool learning, the lack of RMs specifically designed for function-calling tasks has limited progress toward more capable agentic AI. We introduce ToolRM, a family of lightweight generative RMs tailored for general tool-use scenarios. To build these models, we propose a novel pipeline that constructs pairwise preference data using rule-based scoring and multidimensional sampling. This yields ToolPref-Pairwise-30K, a diverse, balanced, and challenging dataset of critique tasks that supports reinforcement learning with verifiable feedback. To evaluate tool-use RMs, we also introduce TRBench$_{BFCL}$, a benchmark built on the agentic evaluation suite BFCL. Trained on our constructed data, models from the Qwen3-4B/8B series achieve up to 14.28% higher accuracy, substantially outperforming frontier models such as Claude 4 and OpenAI o3 in pairwise reward judgments. Beyond training objectives, ToolRM generalizes to broader critique tasks, including Best-of-N sampling and self-correction. Experiments on ACEBench highlight its effectiveness and efficiency, enabling inference-time scaling and reducing output token usage by over 66%. We release data and model checkpoints to facilitate future research.

☆ Learning to Manage Investment Portfolios beyond Simple Utility Functions

Maarten P. Scholl, Mahmoud Mahfouz, Anisoara Calinescu, J. Doyne Farmer

While investment funds publicly disclose their objectives in broad terms, their managers optimize for complex combinations of competing goals that go beyond simple risk-return trade-offs. Traditional approaches attempt to model this through multi-objective utility functions, but face fundamental challenges in specification and parameterization. We propose a generative framework that learns latent representations of fund manager strategies without requiring explicit utility specification. Our approach directly models the conditional probability of a fund's portfolio weights, given stock characteristics, historical returns, previous weights, and a latent variable representing the fund's strategy. Unlike methods based on reinforcement learning or imitation learning, which require specified rewards or labeled expert objectives, our GAN-based architecture learns directly from the joint distribution of observed holdings and market data. We validate our framework on a dataset of 1436 U.S. equity mutual funds. The learned representations successfully capture known investment styles, such as "growth" and "value," while also revealing implicit manager objectives. For instance, we find that while many funds exhibit characteristics of Markowitz-like optimization, they do so with heterogeneous realizations for turnover, concentration, and latent factors. To analyze and interpret the end-to-end model, we develop a series of tests that explain the model, and we show that the benchmark's expert labeling are contained in our model's encoding in a linear interpretable way. Our framework provides a data-driven approach for characterizing investment strategies for applications in market simulation, strategy attribution, and regulatory oversight.

comment: 6th ACM International Conference on AI in Finance, November 15-18, 2025, Singapore

☆ Segmentation over Complexity: Evaluating Ensemble and Hybrid Approaches for Anomaly Detection in Industrial Time Series

Emilio Mastriani, Alessandro Costa, Federico Incardona, Kevin Munari, Sebastiano Spinello

In this study, we investigate the effectiveness of advanced feature engineering and hybrid model architectures for anomaly detection in a multivariate industrial time series, focusing on a steam turbine system. We evaluate the impact of change point-derived statistical features, clustering-based substructure representations, and hybrid learning strategies on detection performance. Despite their theoretical appeal, these complex approaches consistently underperformed compared to a simple Random Forest + XGBoost ensemble trained on segmented data. The ensemble achieved an AUC-ROC of 0.976, F1-score of 0.41, and 100% early detection within the defined time window. Our findings highlight that, in scenarios with highly imbalanced and temporally uncertain data, model simplicity combined with optimized segmentation can outperform more sophisticated architectures, offering greater robustness, interpretability, and operational utility.

comment: This paper is currently under review for presentation at the IEEE SAMI 2026 Conference

☆ Bridging the Gap Between Molecule and Textual Descriptions via Substructure-aware Alignment EMNLP 2025

Hyuntae Park, Yeachan Kim, SangKeun Lee

Molecule and text representation learning has gained increasing interest due to its potential for enhancing the understanding of chemical information. However, existing models often struggle to capture subtle differences between molecules and their descriptions, as they lack the ability to learn fine-grained alignments between molecular substructures and chemical phrases. To address this limitation, we introduce MolBridge, a novel molecule-text learning framework based on substructure-aware alignments. Specifically, we augment the original molecule-description pairs with additional alignment signals derived from molecular substructures and chemical phrases. To effectively learn from these enriched alignments, MolBridge employs substructure-aware contrastive learning, coupled with a self-refinement mechanism that filters out noisy alignment signals. Experimental results show that MolBridge effectively captures fine-grained correspondences and outperforms state-of-the-art baselines on a wide range of molecular benchmarks, highlighting the significance of substructure-aware alignment in molecule-text learning.

comment: EMNLP 2025 (main)

☆ MV-MLM: Bridging Multi-View Mammography and Language for Breast Cancer Diagnosis and Risk Prediction ICCV 2025

Shunjie-Fabian Zheng, Hyeonjun Lee, Thijs Kooi, Ali Diba

Large annotated datasets are essential for training robust Computer-Aided Diagnosis (CAD) models for breast cancer detection or risk prediction. However, acquiring such datasets with fine-detailed annotation is both costly and time-consuming. Vision-Language Models (VLMs), such as CLIP, which are pre-trained on large image-text pairs, offer a promising solution by enhancing robustness and data efficiency in medical imaging tasks. This paper introduces a novel Multi-View Mammography and Language Model for breast cancer classification and risk prediction, trained on a dataset of paired mammogram images and synthetic radiology reports. Our MV-MLM leverages multi-view supervision to learn rich representations from extensive radiology data by employing cross-modal self-supervision across image-text pairs. This includes multiple views and the corresponding pseudo-radiology reports. We propose a novel joint visual-textual learning strategy to enhance generalization and accuracy performance over different data types and tasks to distinguish breast tissues or cancer characteristics(calcification, mass) and utilize these patterns to understand mammography images and predict cancer risk. We evaluated our method on both private and publicly available datasets, demonstrating that the proposed model achieves state-of-the-art performance in three classification tasks: (1) malignancy classification, (2) subtype classification, and (3) image-based cancer risk prediction. Furthermore, the model exhibits strong data efficiency, outperforming existing fully supervised or VLM baselines while trained on synthetic text reports and without the need for actual radiology reports.

comment: Accepted to Computer Vision for Automated Medical Diagnosis (CVAMD) Workshop at ICCV 2025

☆ The FM Agent

Annan Li, Chufan Wu, Zengle Ge, Yee Hin Chong, Zhinan Hou, Lizhe Cao, Cheng Ju, Jianmin Wu, Huaiming Li, Haobo Zhang, Shenghao Feng, Mo Zhao, Fengzhi Qiu, Rui Yang, Mengmeng Zhang, Wenyi Zhu, Yingying Sun, Quan Sun, Shunhao Yan, Danyu Liu, Dawei Yin, Dou Shen

Large language models (LLMs) are catalyzing the development of autonomous AI research agents for scientific and engineering discovery. We present FM Agent, a novel and general-purpose multi-agent framework that leverages a synergistic combination of LLM-based reasoning and large-scale evolutionary search to address complex real-world challenges. The core of FM Agent integrates several key innovations: 1) a cold-start initialization phase incorporating expert guidance, 2) a novel evolutionary sampling strategy for iterative optimization, 3) domain-specific evaluators that combine correctness, effectiveness, and LLM-supervised feedback, and 4) a distributed, asynchronous execution infrastructure built on Ray. Demonstrating broad applicability, our system has been evaluated across diverse domains, including operations research, machine learning, GPU kernel optimization, and classical mathematical problems. FM Agent reaches state-of-the-art results autonomously, without human interpretation or tuning -- 1976.3 on ALE-Bench (+5.2\%), 43.56\% on MLE-Bench (+4.0pp), up to 20x speedups on KernelBench, and establishes new state-of-the-art(SOTA) results on several classical mathematical problems. Beyond academic benchmarks, FM Agent shows considerable promise for both large-scale enterprise R\&D workflows and fundamental scientific research, where it can accelerate innovation, automate complex discovery processes, and deliver substantial engineering and scientific advances with broader societal impact.

☆ Reasoning Curriculum: Bootstrapping Broad LLM Reasoning from Math

Bo Pang, Deqian Kong, Silvio Savarese, Caiming Xiong, Yingbo Zhou

Reinforcement learning (RL) can elicit strong reasoning in large language models (LLMs), yet most open efforts focus on math and code. We propose Reasoning Curriculum, a simple two-stage curriculum that first elicits reasoning skills in pretraining-aligned domains such as math, then adapts and refines these skills across other domains via joint RL. Stage 1 performs a brief cold start and then math-only RL with verifiable rewards to develop reasoning skills. Stage 2 runs joint RL on mixed-domain data to transfer and consolidate these skills. The curriculum is minimal and backbone-agnostic, requiring no specialized reward models beyond standard verifiability checks. Evaluated on Qwen3-4B and Llama-3.1-8B over a multi-domain suite, reasoning curriculum yields consistent gains. Ablations and a cognitive-skill analysis indicate that both stages are necessary and that math-first elicitation increases cognitive behaviors important for solving complex problems. Reasoning Curriculum provides a compact, easy-to-adopt recipe for general reasoning.

comment: 9 pages

☆ Beyond Benchmarks: The Economics of AI Inference

Boqin Zhuang, Jiacheng Qiao, Mingqian Liu, Mingxing Yu, Ping Hong, Rui Li, Xiaoxia Song, Xiangjun Xu, Xu Chen, Yaoyao Ma, Yujie Gao

The inference cost of Large Language Models (LLMs) has become a critical factor in determining their commercial viability and widespread adoption. This paper introduces a quantitative ``economics of inference'' framework, treating the LLM inference process as a compute-driven intelligent production activity. We analyze its marginal cost, economies of scale, and quality of output under various performance configurations. Based on empirical data from WiNEval-3.0, we construct the first ``LLM Inference Production Frontier,'' revealing three principles: diminishing marginal cost, diminishing returns to scale, and an optimal cost-effectiveness zone. This paper not only provides an economic basis for model deployment decisions but also lays an empirical foundation for the future market-based pricing and optimization of AI inference resources.

☆ Beyond Synthetic Benchmarks: Evaluating LLM Performance on Real-World Class-Level Code Generation

Musfiqur Rahman, SayedHassan Khatoonabadi, Emad Shihab

Large language models (LLMs) have advanced code generation at the function level, yet their ability to produce correct class-level implementations in authentic software projects remains poorly understood. This work introduces a novel benchmark derived from open-source repositories, comprising real-world classes divided into seen and unseen partitions to evaluate generalization under practical conditions. The evaluation examines multiple LLMs under varied input specifications, retrieval-augmented configurations, and documentation completeness levels. Results reveal a stark performance disparity: LLMs achieve 84% to 89% correctness on established synthetic benchmarks but only 25% to 34% on real-world class tasks, with negligible differences between familiar and novel codebases. Comprehensive docstrings yield modest gains of 1% to 3% in functional accuracy, though statistical significance is rare. Retrieval-augmented generation proves most effective with partial documentation, improving correctness by 4% to 7% by supplying concrete implementation patterns absent from specifications. Error profiling identifies AttributeError, TypeError, and AssertionError as dominant failure modes (84% of cases), with synthetic tests overemphasizing assertion issues and real-world scenarios highlighting type and attribute mismatches. Retrieval augmentation reduces logical flaws but can introduce dependency conflicts. The benchmark and analysis expose critical limitations in current LLM capabilities for class-level engineering, offering actionable insights for enhancing context modelling, documentation strategies, and retrieval integration in production code assistance tools.

comment: Pre-print prepared for journal submission

☆ WOD-E2E: Waymo Open Dataset for End-to-End Driving in Challenging Long-tail Scenarios

Runsheng Xu, Hubert Lin, Wonseok Jeon, Hao Feng, Yuliang Zou, Liting Sun, John Gorman, Kate Tolstaya, Sarah Tang, Brandyn White, Ben Sapp, Mingxing Tan, Jyh-Jing Hwang, Drago Anguelov

Vision-based end-to-end (E2E) driving has garnered significant interest in the research community due to its scalability and synergy with multimodal large language models (MLLMs). However, current E2E driving benchmarks primarily feature nominal scenarios, failing to adequately test the true potential of these systems. Furthermore, existing open-loop evaluation metrics often fall short in capturing the multi-modal nature of driving or effectively evaluating performance in long-tail scenarios. To address these gaps, we introduce the Waymo Open Dataset for End-to-End Driving (WOD-E2E). WOD-E2E contains 4,021 driving segments (approximately 12 hours), specifically curated for challenging long-tail scenarios that that are rare in daily life with an occurring frequency of less than 0.03%. Concretely, each segment in WOD-E2E includes the high-level routing information, ego states, and 360-degree camera views from 8 surrounding cameras. To evaluate the E2E driving performance on these long-tail situations, we propose a novel open-loop evaluation metric: Rater Feedback Score (RFS). Unlike conventional metrics that measure the distance between predicted way points and the logs, RFS measures how closely the predicted trajectory matches rater-annotated trajectory preference labels. We have released rater preference labels for all WOD-E2E validation set segments, while the held out test set labels have been used for the 2025 WOD-E2E Challenge. Through our work, we aim to foster state of the art research into generalizable, robust, and safe end-to-end autonomous driving agents capable of handling complex real-world situations.

☆ EgoExo-Con: Exploring View-Invariant Video Temporal Understanding

Minjoon Jung, Junbin Xiao, Junghyun Kim, Byoung-Tak Zhang, Angela Yao

Can Video-LLMs achieve consistent temporal understanding when videos capture the same event from different viewpoints? To study this, we introduce EgoExo-Con (Consistency), a benchmark of comprehensively synchronized egocentric and exocentric video pairs with human-refined queries in natural language. EgoExo-Con emphasizes two temporal understanding tasks: Temporal Verification and Temporal Grounding. It evaluates not only correctness but consistency across viewpoints. Our analysis reveals two critical limitations of existing Video-LLMs: (1) models often fail to maintain consistency, with results far worse than their single-view performances. (2) When naively finetuned with synchronized videos of both viewpoints, the models show improved consistency but often underperform those trained on a single view. For improvements, we propose View-GRPO, a novel reinforcement learning framework that effectively strengthens view-specific temporal reasoning while encouraging consistent comprehension across viewpoints. Our method demonstrates its superiority over naive SFT and GRPO, especially for improving cross-view consistency. All resources will be made publicly available.

comment: project page: \url{https://minjoong507.github.io/projects/EgoExo-Con/}

☆ Security Risk of Misalignment between Text and Image in Multi-modal Model

Xiaosen Wang, Zhijin Ge, Shaokang Wang

Despite the notable advancements and versatility of multi-modal diffusion models, such as text-to-image models, their susceptibility to adversarial inputs remains underexplored. Contrary to expectations, our investigations reveal that the alignment between textual and Image modalities in existing diffusion models is inadequate. This misalignment presents significant risks, especially in the generation of inappropriate or Not-Safe-For-Work (NSFW) content. To this end, we propose a novel attack called Prompt-Restricted Multi-modal Attack (PReMA) to manipulate the generated content by modifying the input image in conjunction with any specified prompt, without altering the prompt itself. PReMA is the first attack that manipulates model outputs by solely creating adversarial images, distinguishing itself from prior methods that primarily generate adversarial prompts to produce NSFW content. Consequently, PReMA poses a novel threat to the integrity of multi-modal diffusion models, particularly in image-editing applications that operate with fixed prompts. Comprehensive evaluations conducted on image inpainting and style transfer tasks across various models confirm the potent efficacy of PReMA.

☆ SAFE: A Novel Approach to AI Weather Evaluation through Stratified Assessments of Forecasts over Earth

Nick Masi, Randall Balestriero

The dominant paradigm in machine learning is to assess model performance based on average loss across all samples in some test set. This amounts to averaging performance geospatially across the Earth in weather and climate settings, failing to account for the non-uniform distribution of human development and geography. We introduce Stratified Assessments of Forecasts over Earth (SAFE), a package for elucidating the stratified performance of a set of predictions made over Earth. SAFE integrates various data domains to stratify by different attributes associated with geospatial gridpoints: territory (usually country), global subregion, income, and landcover (land or water). This allows us to examine the performance of models for each individual stratum of the different attributes (e.g., the accuracy in every individual country). To demonstrate its importance, we utilize SAFE to benchmark a zoo of state-of-the-art AI-based weather prediction models, finding that they all exhibit disparities in forecasting skill across every attribute. We use this to seed a benchmark of model forecast fairness through stratification at different lead times for various climatic variables. By moving beyond globally-averaged metrics, we for the first time ask: where do models perform best or worst, and which models are most fair? To support further work in this direction, the SAFE package is open source and available at https://github.com/N-Masi/safe

☆ GUI Knowledge Bench: Revealing the Knowledge Gap Behind VLM Failures in GUI Tasks

Chenrui Shi, Zedong Yu, Zhi Gao, Ruining Feng, Enqi Liu, Yuwei Wu, Yunde Jia, Liuyu Xiang, Zhaofeng He, Qing Li

Large vision language models (VLMs) have advanced graphical user interface (GUI) task automation but still lag behind humans. We hypothesize this gap stems from missing core GUI knowledge, which existing training schemes (such as supervised fine tuning and reinforcement learning) alone cannot fully address. By analyzing common failure patterns in GUI task execution, we distill GUI knowledge into three dimensions: (1) interface perception, knowledge about recognizing widgets and system states; (2) interaction prediction, knowledge about reasoning action state transitions; and (3) instruction understanding, knowledge about planning, verifying, and assessing task completion progress. We further introduce GUI Knowledge Bench, a benchmark with multiple choice and yes/no questions across six platforms (Web, Android, MacOS, Windows, Linux, IOS) and 292 applications. Our evaluation shows that current VLMs identify widget functions but struggle with perceiving system states, predicting actions, and verifying task completion. Experiments on real world GUI tasks further validate the close link between GUI knowledge and task success. By providing a structured framework for assessing GUI knowledge, our work supports the selection of VLMs with greater potential prior to downstream training and provides insights for building more capable GUI agents.

☆ Lean4Physics: Comprehensive Reasoning Framework for College-level Physics in Lean4

Yuxin Li, Minghao Liu, Ruida Wang, Wenzhao Ji, Zhitao He, Rui Pan, Junming Huang, Tong Zhang, Yi R. Fung

We present **Lean4PHYS**, a comprehensive reasoning framework for college-level physics problems in Lean4. **Lean4PHYS** includes *LeanPhysBench*, a college-level benchmark for formal physics reasoning in Lean4, which contains 200 hand-crafted and peer-reviewed statements derived from university textbooks and physics competition problems. To establish a solid foundation for formal reasoning in physics, we also introduce *PhysLib*, a community-driven repository containing fundamental unit systems and theorems essential for formal physics reasoning. Based on the benchmark and Lean4 repository we composed in **Lean4PHYS**, we report baseline results using major expert Math Lean4 provers and state-of-the-art closed-source models, with the best performance of DeepSeek-Prover-V2-7B achieving only 16% and Claude-Sonnet-4 achieving 35%. We also conduct a detailed analysis showing that our *PhysLib* can achieve an average improvement of 11.75% in model performance. This demonstrates the challenging nature of our *LeanPhysBench* and the effectiveness of *PhysLib*. To the best of our knowledge, this is the first study to provide a physics benchmark in Lean4.

☆ Network-Constrained Policy Optimization for Adaptive Multi-agent Vehicle Routing

Fazel Arasteh, Arian Haghparast, Manos Papagelis

Traffic congestion in urban road networks leads to longer trip times and higher emissions, especially during peak periods. While the Shortest Path First (SPF) algorithm is optimal for a single vehicle in a static network, it performs poorly in dynamic, multi-vehicle settings, often worsening congestion by routing all vehicles along identical paths. We address dynamic vehicle routing through a multi-agent reinforcement learning (MARL) framework for coordinated, network-aware fleet navigation. We first propose Adaptive Navigation (AN), a decentralized MARL model where each intersection agent provides routing guidance based on (i) local traffic and (ii) neighborhood state modeled using Graph Attention Networks (GAT). To improve scalability in large networks, we further propose Hierarchical Hub-based Adaptive Navigation (HHAN), an extension of AN that assigns agents only to key intersections (hubs). Vehicles are routed hub-to-hub under agent control, while SPF handles micro-routing within each hub region. For hub coordination, HHAN adopts centralized training with decentralized execution (CTDE) under the Attentive Q-Mixing (A-QMIX) framework, which aggregates asynchronous vehicle decisions via attention. Hub agents use flow-aware state features that combine local congestion and predictive dynamics for proactive routing. Experiments on synthetic grids and real urban maps (Toronto, Manhattan) show that AN reduces average travel time versus SPF and learning baselines, maintaining 100% routing success. HHAN scales to networks with hundreds of intersections, achieving up to 15.9% improvement under heavy traffic. These findings highlight the potential of network-constrained MARL for scalable, coordinated, and congestion-aware routing in intelligent transportation systems.

comment: 29 pages, 12 figures. Fazel Arasteh and Arian Haghparast contributed equally to this research. Submitted to ACM Transactions on Spatial Algorithms and Systems (TSAS). The code for this work is publicly available at https://github.com/Arianhgh/HHAN

☆ Nirvana: A Specialized Generalist Model With Task-Aware Memory Mechanism

Yuhua Jiang, Shuang Cheng, Yihao Liu, Ermo Hua, Che Jiang, Weigao Sun, Yu Cheng, Feifei Gao, Biqing Qi, Bowen Zhou

Specialized Generalist Models (SGMs) aim to preserve broad capabilities while achieving expert-level performance in target domains. However, traditional LLM structures including Transformer, Linear Attention, and hybrid models do not employ specialized memory mechanism guided by task information. In this paper, we present Nirvana, an SGM with specialized memory mechanism, linear time complexity, and test-time task information extraction. Besides, we propose the Task-Aware Memory Trigger ($\textit{Trigger}$) that flexibly adjusts memory mechanism based on the current task's requirements. In Trigger, each incoming sample is treated as a self-supervised fine-tuning task, enabling Nirvana to adapt its task-related parameters on the fly to domain shifts. We also design the Specialized Memory Updater ($\textit{Updater}$) that dynamically memorizes the context guided by Trigger. We conduct experiments on both general language tasks and specialized medical tasks. On a variety of natural language modeling benchmarks, Nirvana achieves competitive or superior results compared to the existing LLM structures. To prove the effectiveness of Trigger on specialized tasks, we test Nirvana's performance on a challenging medical task, i.e., Magnetic Resonance Imaging (MRI). We post-train frozen Nirvana backbone with lightweight codecs on paired electromagnetic signals and MRI images. Despite the frozen Nirvana backbone, Trigger guides the model to adapt to the MRI domain with the change of task-related parameters. Nirvana achieves higher-quality MRI reconstruction compared to conventional MRI models as well as the models with traditional LLMs' backbone, and can also generate accurate preliminary clinical reports accordingly.

☆ Learning Geometry: A Framework for Building Adaptive Manifold Models through Metric Optimization

Di Zhang

This paper proposes a novel paradigm for machine learning that moves beyond traditional parameter optimization. Unlike conventional approaches that search for optimal parameters within a fixed geometric space, our core idea is to treat the model itself as a malleable geometric entity. Specifically, we optimize the metric tensor field on a manifold with a predefined topology, thereby dynamically shaping the geometric structure of the model space. To achieve this, we construct a variational framework whose loss function carefully balances data fidelity against the intrinsic geometric complexity of the manifold. The former ensures the model effectively explains observed data, while the latter acts as a regularizer, penalizing overly curved or irregular geometries to encourage simpler models and prevent overfitting. To address the computational challenges of this infinite-dimensional optimization problem, we introduce a practical method based on discrete differential geometry: the continuous manifold is discretized into a triangular mesh, and the metric tensor is parameterized by edge lengths, enabling efficient optimization using automatic differentiation tools. Theoretical analysis reveals a profound analogy between our framework and the Einstein-Hilbert action in general relativity, providing an elegant physical interpretation for the concept of "data-driven geometry". We further argue that even with fixed topology, metric optimization offers significantly greater expressive power than models with fixed geometry. This work lays a solid foundation for constructing fully dynamic "meta-learners" capable of autonomously evolving their geometry and topology, and it points to broad application prospects in areas such as scientific model discovery and robust representation learning.

comment: 9 pages

☆ Data-driven Projection Generation for Efficiently Solving Heterogeneous Quadratic Programming Problems

Tomoharu Iwata, Futoshi Futami

We propose a data-driven framework for efficiently solving quadratic programming (QP) problems by reducing the number of variables in high-dimensional QPs using instance-specific projection. A graph neural network-based model is designed to generate projections tailored to each QP instance, enabling us to produce high-quality solutions even for previously unseen problems. The model is trained on heterogeneous QPs to minimize the expected objective value evaluated on the projected solutions. This is formulated as a bilevel optimization problem; the inner optimization solves the QP under a given projection using a QP solver, while the outer optimization updates the model parameters. We develop an efficient algorithm to solve this bilevel optimization problem, which computes parameter gradients without backpropagating through the solver. We provide a theoretical analysis of the generalization ability of solving QPs with projection matrices generated by neural networks. Experimental results demonstrate that our method produces high-quality feasible solutions with reduced computation time, outperforming existing methods.

☆ Can AI be Accountable?

Andrew L. Kun

The AI we use is powerful, and its power is increasing rapidly. If this powerful AI is to serve the needs of consumers, voters, and decision makers, then it is imperative that the AI is accountable. In general, an agent is accountable to a forum if the forum can request information from the agent about its actions, if the forum and the agent can discuss this information, and if the forum can sanction the agent. Unfortunately, in too many cases today's AI is not accountable -- we cannot question it, enter into a discussion with it, let alone sanction it. In this chapter we relate the general definition of accountability to AI, we illustrate what it means for AI to be accountable and unaccountable, and we explore approaches that can improve our chances of living in a world where all AI is accountable to those who are affected by it.

comment: To be published as a chapter in Daniele Quercia and Marios Constantinides (Eds.). Operationalizing Responsible AI. Cambridge University Press. Forthcoming

☆ Dynamic VLM-Guided Negative Prompting for Diffusion Models NeurIPS 2025

Hoyeon Chang, Seungjin Kim, Yoonseok Choi

We propose a novel approach for dynamic negative prompting in diffusion models that leverages Vision-Language Models (VLMs) to adaptively generate negative prompts during the denoising process. Unlike traditional Negative Prompting methods that use fixed negative prompts, our method generates intermediate image predictions at specific denoising steps and queries a VLM to produce contextually appropriate negative prompts. We evaluate our approach on various benchmark datasets and demonstrate the trade-offs between negative guidance strength and text-image alignment.

comment: 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: The First Workshop on Generative and Protective AI for Content Creation

☆ Do Students Debias Like Teachers? On the Distillability of Bias Mitigation Methods

Jiali Cheng, Chirag Agarwal, Hadi Amiri

Knowledge distillation (KD) is an effective method for model compression and transferring knowledge between models. However, its effect on model's robustness against spurious correlations that degrade performance on out-of-distribution data remains underexplored. This study investigates the effect of knowledge distillation on the transferability of ``debiasing'' capabilities from teacher models to student models on natural language inference (NLI) and image classification tasks. Through extensive experiments, we illustrate several key findings: (i) overall the debiasing capability of a model is undermined post-KD; (ii) training a debiased model does not benefit from injecting teacher knowledge; (iii) although the overall robustness of a model may remain stable post-distillation, significant variations can occur across different types of biases; and (iv) we pin-point the internal attention pattern and circuit that causes the distinct behavior post-KD. Given the above findings, we propose three effective solutions to improve the distillability of debiasing methods: developing high quality data for augmentation, implementing iterative knowledge distillation, and initializing student models with weights obtained from teacher models. To the best of our knowledge, this is the first study on the effect of KD on debiasing and its interenal mechanism at scale. Our findings provide understandings on how KD works and how to design better debiasing methods.

☆ SIRAJ: Diverse and Efficient Red-Teaming for LLM Agents via Distilled Structured Reasoning

Kaiwen Zhou, Ahmed Elgohary, A S M Iftekhar, Amin Saied

The ability of LLM agents to plan and invoke tools exposes them to new safety risks, making a comprehensive red-teaming system crucial for discovering vulnerabilities and ensuring their safe deployment. We present SIRAJ: a generic red-teaming framework for arbitrary black-box LLM agents. We employ a dynamic two-step process that starts with an agent definition and generates diverse seed test cases that cover various risk outcomes, tool-use trajectories, and risk sources. Then, it iteratively constructs and refines model-based adversarial attacks based on the execution trajectories of former attempts. To optimize the red-teaming cost, we present a model distillation approach that leverages structured forms of a teacher model's reasoning to train smaller models that are equally effective. Across diverse evaluation agent settings, our seed test case generation approach yields 2 -- 2.5x boost to the coverage of risk outcomes and tool-calling trajectories. Our distilled 8B red-teamer model improves attack success rate by 100%, surpassing the 671B Deepseek-R1 model. Our ablations and analyses validate the effectiveness of the iterative framework, structured reasoning, and the generalization of our red-teamer models.

☆ Artificial Intelligence-Enabled Analysis of Radiology Reports: Epidemiology and Consequences of Incidental Thyroid Findings

Felipe Larios, Mariana Borras-Osorio, Yuqi Wu, Ana Gabriela Claros, David Toro-Tobon, Esteban Cabezas, Ricardo Loor-Torres, Maria Mateo Chavez, Kerly Guevara Maldonado, Luis Vilatuna Andrango, Maria Lizarazo Jimenez, Ivan Mateo Alzamora, Misk Al Zahidy, Marcelo Montero, Ana Cristina Proano, Cristian Soto Jacome, Jungwei W. Fan, Oscar J. Ponce-Ponte, Megan E. Branda, Naykky Singh Ospina, Juan P. Brito

Importance Incidental thyroid findings (ITFs) are increasingly detected on imaging performed for non-thyroid indications. Their prevalence, features, and clinical consequences remain undefined. Objective To develop, validate, and deploy a natural language processing (NLP) pipeline to identify ITFs in radiology reports and assess their prevalence, features, and clinical outcomes. Design, Setting, and Participants Retrospective cohort of adults without prior thyroid disease undergoing thyroid-capturing imaging at Mayo Clinic sites from July 1, 2017, to September 30, 2023. A transformer-based NLP pipeline identified ITFs and extracted nodule characteristics from image reports from multiple modalities and body regions. Main Outcomes and Measures Prevalence of ITFs, downstream thyroid ultrasound, biopsy, thyroidectomy, and thyroid cancer diagnosis. Logistic regression identified demographic and imaging-related factors. Results Among 115,683 patients (mean age, 56.8 [SD 17.2] years; 52.9% women), 9,077 (7.8%) had an ITF, of which 92.9% were nodules. ITFs were more likely in women, older adults, those with higher BMI, and when imaging was ordered by oncology or internal medicine. Compared with chest CT, ITFs were more likely via neck CT, PET, and nuclear medicine scans. Nodule characteristics were poorly documented, with size reported in 44% and other features in fewer than 15% (e.g. calcifications). Compared with patients without ITFs, those with ITFs had higher odds of thyroid nodule diagnosis, biopsy, thyroidectomy and thyroid cancer diagnosis. Most cancers were papillary, and larger when detected after ITFs vs no ITF. Conclusions ITFs were common and strongly associated with cascades leading to the detection of small, low-risk cancers. These findings underscore the role of ITFs in thyroid cancer overdiagnosis and the need for standardized reporting and more selective follow-up.

♻ ☆ UniSite: The First Cross-Structure Dataset and Learning Framework for End-to-End Ligand Binding Site Detection

Jigang Fan, Quanlin Wu, Shengjie Luo, Liwei Wang

The detection of ligand binding sites for proteins is a fundamental step in Structure-Based Drug Design. Despite notable advances in recent years, existing methods, datasets, and evaluation metrics are confronted with several key challenges: (1) current datasets and methods are centered on individual protein-ligand complexes and neglect that diverse binding sites may exist across multiple complexes of the same protein, introducing significant statistical bias; (2) ligand binding site detection is typically modeled as a discontinuous workflow, employing binary segmentation and subsequent clustering algorithms; (3) traditional evaluation metrics do not adequately reflect the actual performance of different binding site prediction methods. To address these issues, we first introduce UniSite-DS, the first UniProt (Unique Protein)-centric ligand binding site dataset, which contains 4.81 times more multi-site data and 2.08 times more overall data compared to the previously most widely used datasets. We then propose UniSite, the first end-to-end ligand binding site detection framework supervised by set prediction loss with bijective matching. In addition, we introduce Average Precision based on Intersection over Union (IoU) as a more accurate evaluation metric for ligand binding site prediction. Extensive experiments on UniSite-DS and several representative benchmark datasets demonstrate that IoU-based Average Precision provides a more accurate reflection of prediction quality, and that UniSite outperforms current state-of-the-art methods in ligand binding site detection. The dataset and codes will be made publicly available at https://github.com/quanlin-wu/unisite.

♻ ☆ Completion $\neq$ Collaboration: Scaling Collaborative Effort with Agents

Shannon Zejiang Shen, Valerie Chen, Ken Gu, Alexis Ross, Zixian Ma, Jillian Ross, Alex Gu, Chenglei Si, Wayne Chi, Andi Peng, Jocelyn J Shen, Ameet Talwalkar, Tongshuang Wu, David Sontag

Current evaluations of agents remain centered around one-shot task completion, failing to account for the inherently iterative and collaborative nature of many real-world problems, where human goals are often underspecified and evolve. We argue for a shift from building and assessing task completion agents to developing collaborative agents, assessed not only by the quality of their final outputs but by how well they engage with and enhance human effort throughout the problem-solving process. To support this shift, we introduce collaborative effort scaling, a framework that captures how an agent's utility grows with increasing user involvement. Through case studies and simulated evaluations, we show that state-of-the-art agents often underperform in multi-turn, real-world scenarios, revealing a missing ingredient in agent design: the ability to sustain engagement and scaffold user understanding. Collaborative effort scaling offers a lens for diagnosing agent behavior and guiding development toward more effective interactions.

comment: 22 pages, 5 figures, 3 tables

♻ ☆ Controlling Thinking Speed in Reasoning Models NeurIPS 2025

Zhengkai Lin, Zhihang Fu, Ze Chen, Chao Chen, Liang Xie, Wenxiao Wang, Deng Cai, Zheng Wang, Jieping Ye

Human cognition is theorized to operate in two modes: fast, intuitive System 1 thinking and slow, deliberate System 2 thinking. While current Large Reasoning Models (LRMs) excel at System 2 thinking, their inability to perform fast thinking leads to high computational overhead and latency. In this work, we enable LRMs to approximate human intelligence through dynamic thinking speed adjustment, optimizing accuracy-efficiency trade-offs. Our approach addresses two key questions: (1) how to control thinking speed in LRMs, and (2) when to adjust it for optimal performance. For the first question, we identify the steering vector that governs slow-fast thinking transitions in LRMs' representation space. Using this vector, we achieve the first representation editing-based test-time scaling effect, outperforming existing prompt-based scaling methods. For the second question, we apply real-time difficulty estimation to signal reasoning segments of varying complexity. Combining these techniques, we propose the first reasoning strategy that enables fast processing of easy steps and deeper analysis for complex reasoning. Without any training or additional cost, our plug-in module delivers an average +1.3% accuracy with -8.6% token usage across leading LRMs and advanced reasoning benchmarks. All of our algorithms are implemented based on vLLM and are expected to support broader applications and inspire future research.

comment: NeurIPS 2025 Spotlight

♻ ☆ RLBFF: Binary Flexible Feedback to bridge between Human Feedback & Verifiable Rewards

Zhilin Wang, Jiaqi Zeng, Olivier Delalleau, Ellie Evans, Daniel Egert, Hoo-Chang Shin, Felipe Soares, Yi Dong, Oleksii Kuchaiev

Reinforcement Learning with Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) are the main RL paradigms used in LLM post-training, each offering distinct advantages. However, RLHF struggles with interpretability and reward hacking because it relies on human judgments that usually lack explicit criteria, whereas RLVR is limited in scope by its focus on correctness-based verifiers. We propose Reinforcement Learning with Binary Flexible Feedback (RLBFF), which combines the versatility of human-driven preferences with the precision of rule-based verification, enabling reward models to capture nuanced aspects of response quality beyond mere correctness. RLBFF extracts principles that can be answered in a binary fashion (e.g. accuracy of information: yes, or code readability: no) from natural language feedback. Such principles can then be used to ground Reward Model training as an entailment task (response satisfies or does not satisfy an arbitrary principle). We show that Reward Models trained in this manner can outperform Bradley-Terry models when matched for data and achieve top performance on RM-Bench (86.2%) and JudgeBench (81.4%, #1 on leaderboard as of September 24, 2025). Additionally, users can specify principles of interest at inference time to customize the focus of our reward models, in contrast to Bradley-Terry models. Finally, we present a fully open source recipe (including data) to align Qwen3-32B using RLBFF and our Reward Model, to match or exceed the performance of o3-mini and DeepSeek R1 on general alignment benchmarks of MT-Bench, WildBench, and Arena Hard v2 (at <5% of the inference cost). Models: https://huggingface.co/collections/nvidia/reward-models-10-2025

comment: Added link to access models: https://huggingface.co/collections/nvidia/reward-models-10-2025

♻ ☆ Curriculum Abductive Learning NeurIPS 2025

Wen-Chao Hu, Qi-Jie Li, Lin-Han Jia, Cunjing Ge, Yu-Feng Li, Yuan Jiang, Zhi-Hua Zhou

Abductive Learning (ABL) integrates machine learning with logical reasoning in a loop: a learning model predicts symbolic concept labels from raw inputs, which are revised through abduction using domain knowledge and then fed back for retraining. However, due to the nondeterminism of abduction, the training process often suffers from instability, especially when the knowledge base is large and complex, resulting in a prohibitively large abduction space. While prior works focus on improving candidate selection within this space, they typically treat the knowledge base as a static black box. In this work, we propose Curriculum Abductive Learning (C-ABL), a method that explicitly leverages the internal structure of the knowledge base to address the ABL training challenges. C-ABL partitions the knowledge base into a sequence of sub-bases, progressively introduced during training. This reduces the abduction space throughout training and enables the model to incorporate logic in a stepwise, smooth way. Experiments across multiple tasks show that C-ABL outperforms previous ABL implementations, significantly improves training stability, convergence speed, and final accuracy, especially under complex knowledge setting.

comment: Accepted by NeurIPS 2025, 22 pages, 6 figures

♻ ☆ Guided Model Merging for Hybrid Data Learning: Leveraging Centralized Data to Refine Decentralized Models WACV 2026

Junyi Zhu, Ruicong Yao, Taha Ceritli, Savas Ozkan, Matthew B. Blaschko, Eunchung Noh, Jeongwon Min, Cho Jung Min, Mete Ozay

Current network training paradigms primarily focus on either centralized or decentralized data regimes. However, in practice, data availability often exhibits a hybrid nature, where both regimes coexist. This hybrid setting presents new opportunities for model training, as the two regimes offer complementary trade-offs: decentralized data is abundant but subject to heterogeneity and communication constraints, while centralized data, though limited in volume and potentially unrepresentative, enables better curation and high-throughput access. Despite its potential, effectively combining these paradigms remains challenging, and few frameworks are tailored to hybrid data regimes. To address this, we propose a novel framework that constructs a model atlas from decentralized models and leverages centralized data to refine a global model within this structured space. The refined model is then used to reinitialize the decentralized models. Our method synergizes federated learning (to exploit decentralized data) and model merging (to utilize centralized data), enabling effective training under hybrid data availability. Theoretically, we show that our approach achieves faster convergence than methods relying solely on decentralized data, due to variance reduction in the merging process. Extensive experiments demonstrate that our framework consistently outperforms purely centralized, purely decentralized, and existing hybrid-adaptable methods. Notably, our method remains robust even when the centralized and decentralized data domains differ or when decentralized data contains noise, significantly broadening its applicability.

comment: Accepted at WACV 2026

♻ ☆ Refine-n-Judge: Curating High-Quality Preference Chains for LLM-Fine-Tuning

Derin Cayir, Renjie Tao, Rashi Rungta, Kai Sun, Sean Chen, Haidar Khan, Minseok Kim, Julia Reinspach, Yue Liu

Large Language Models (LLMs) have demonstrated remarkable progress through preference-based fine-tuning, which critically depends on the quality of the underlying training data. While human feedback is essential for improving data quality, it is costly and does not scale well. In this paper, we introduce Refine-n-Judge, an automated iterative approach that leverages a single LLM as both a refiner and a judge to enhance dataset quality. Unlike existing iterative refinement methods, Refine-n-Judge employs an LLM to both generate refinements and explicitly evaluate each improvement, ensuring that every iteration meaningfully enhances the dataset without requiring additional human annotation or a separate reward model. At each step, the LLM refines a response and judges whether the refinement is an improvement over the previous answer. This process continues until the LLM prefers the initial answer over the refinement, indicating no further improvements. This produces sequences of increasing quality, preference-labeled responses ideal for fine-tuning. We demonstrate the effectiveness of Refine-n-Judge across a range of public datasets spanning five corpora, targeting tasks such as coding, math, and conversation. Models (Llama 3.1-8B and Llama 3.3-70B) fine-tuned on Refine-n-Judge-enhanced datasets were preferred by LLM judges in over 74% of comparisons against models tuned on the original dataset by GPT-4. Additionally, we report performance gains: +5% on AlpacaEval and AlpacaEval 2.0, and +19% on MT-Bench. Our results indicate that Refine-n-Judge produces high-quality datasets and scalable model improvements.

♻ ☆ CompoST: A Benchmark for Analyzing the Ability of LLMs To Compositionally Interpret Questions in a QALD Setting ISWC 2025

David Maria Schmidt, Raoul Schubert, Philipp Cimiano

Language interpretation is a compositional process, in which the meaning of more complex linguistic structures is inferred from the meaning of their parts. Large language models possess remarkable language interpretation capabilities and have been successfully applied to interpret questions by mapping them to SPARQL queries. An open question is how systematic this interpretation process is. Toward this question, in this paper, we propose a benchmark for investigating to what extent the abilities of LLMs to interpret questions are actually compositional. For this, we generate three datasets of varying difficulty based on graph patterns in DBpedia, relying on Lemon lexica for verbalization. Our datasets are created in a very controlled fashion in order to test the ability of LLMs to interpret structurally complex questions, given that they have seen the atomic building blocks. This allows us to evaluate to what degree LLMs are able to interpret complex questions for which they "understand" the atomic parts. We conduct experiments with models of different sizes using both various prompt and few-shot optimization techniques as well as fine-tuning. Our results show that performance in terms of macro $F_1$ degrades from $0.45$ over $0.26$ down to $0.09$ with increasing deviation from the samples optimized on. Even when all necessary information was provided to the model in the input, the $F_1$ scores do not exceed $0.57$ for the dataset of lowest complexity. We thus conclude that LLMs struggle to systematically and compositionally interpret questions and map them into SPARQL queries.

comment: Research Track, 24th International Semantic Web Conference (ISWC 2025), November 2-6, 2025, Nara, Japan

♻ ☆ Detecting Early and Implicit Suicidal Ideation via Longitudinal and Information Environment Signals on Social Media

Soorya Ram Shimgekar, Ruining Zhao, Agam Goyal, Violeta J. Rodriguez, Paul A. Bloom, Hari Sundaram, Koustuv Saha

On social media, many individuals experiencing suicidal ideation (SI) do not disclose their distress explicitly. Instead, signs may surface indirectly through everyday posts or peer interactions. Detecting such implicit signals early is critical but remains challenging. We frame early and implicit SI as a forward-looking prediction task and develop a computational framework that models a user's information environment, consisting of both their longitudinal posting histories as well as the discourse of their socially proximal peers. We adopted a composite network centrality measure to identify top neighbors of a user, and temporally aligned the user's and neighbors' interactions -- integrating the multi-layered signals in a fine-tuned DeBERTa-v3 model. In a Reddit study of 1,000 (500 Case and 500 Control) users, our approach improves early and implicit SI detection by 15% over individual-only baselines. These findings highlight that peer interactions offer valuable predictive signals and carry broader implications for designing early detection systems that capture indirect as well as masked expressions of risk in online environments.

♻ ☆ Audio Signal Processing Using Time Domain Mel-Frequency Wavelet Coefficient

Rinku Sebastian, Simon O'Keefe, Martin Trefzer

Extracting features from the speech is the most critical process in speech signal processing. Mel Frequency Cepstral Coefficients (MFCC) are the most widely used features in the majority of the speaker and speech recognition applications, as the filtering in this feature is similar to the filtering taking place in the human ear. But the main drawback of this feature is that it provides only the frequency information of the signal but does not provide the information about at what time which frequency is present. The wavelet transform, with its flexible time-frequency window, provides time and frequency information of the signal and is an appropriate tool for the analysis of non-stationary signals like speech. On the other hand, because of its uniform frequency scaling, a typical wavelet transform may be less effective in analysing speech signals, have poorer frequency resolution in low frequencies, and be less in line with human auditory perception. Hence, it is necessary to develop a feature that incorporates the merits of both MFCC and wavelet transform. A great deal of studies are trying to combine both these features. The present Wavelet Transform based Mel-scaled feature extraction methods require more computation when a wavelet transform is applied on top of Mel-scale filtering, since it adds extra processing steps. Here we are proposing a method to extract Mel scale features in time domain combining the concept of wavelet transform, thus reducing the computational burden of time-frequency conversion and the complexity of wavelet extraction. Combining our proposed Time domain Mel frequency Wavelet Coefficient(TMFWC) technique with the reservoir computing methodology has significantly improved the efficiency of audio signal processing.

♻ ☆ MaskCaptioner: Learning to Jointly Segment and Caption Object Trajectories in Videos

Gabriel Fiastre, Antoine Yang, Cordelia Schmid

Dense Video Object Captioning (DVOC) is the task of jointly detecting, tracking, and captioning object trajectories in a video, requiring the ability to understand spatio-temporal details and describe them in natural language. Due to the complexity of the task and the high cost associated with manual annotation, previous approaches resort to disjoint training strategies, potentially leading to suboptimal performance. To circumvent this issue, we propose to generate captions about spatio-temporally localized entities leveraging a state-of-the-art VLM. By extending the LVIS and LV-VIS datasets with our synthetic captions (LVISCap and LV-VISCap), we train MaskCaptioner, an end-to-end model capable of jointly detecting, segmenting, tracking and captioning object trajectories. Moreover, with pretraining on LVISCap and LV-VISCap, MaskCaptioner achieves state-of-the-art DVOC results on three existing benchmarks, VidSTG, VLN and BenSMOT. The datasets and code are available at https://www.gabriel.fiastre.fr/maskcaptioner/.

comment: 20 pages, 8 figures

♻ ☆ LatentBreak: Jailbreaking Large Language Models through Latent Space Feedback

Raffaele Mura, Giorgio Piras, Kamilė Lukošiūtė, Maura Pintor, Amin Karbasi, Battista Biggio

Jailbreaks are adversarial attacks designed to bypass the built-in safety mechanisms of large language models. Automated jailbreaks typically optimize an adversarial suffix or adapt long prompt templates by forcing the model to generate the initial part of a restricted or harmful response. In this work, we show that existing jailbreak attacks that leverage such mechanisms to unlock the model response can be detected by a straightforward perplexity-based filtering on the input prompt. To overcome this issue, we propose LatentBreak, a white-box jailbreak attack that generates natural adversarial prompts with low perplexity capable of evading such defenses. LatentBreak substitutes words in the input prompt with semantically-equivalent ones, preserving the initial intent of the prompt, instead of adding high-perplexity adversarial suffixes or long templates. These words are chosen by minimizing the distance in the latent space between the representation of the adversarial prompt and that of harmless requests. Our extensive evaluation shows that LatentBreak leads to shorter and low-perplexity prompts, thus outperforming competing jailbreak algorithms against perplexity-based filters on multiple safety-aligned models.

♻ ☆ SignalLLM: A General-Purpose LLM Agent Framework for Automated Signal Processing

Junlong Ke, Qiying Hu, Shenghai Yuan, Yuecong Xu, Jianfei Yang

Modern signal processing (SP) pipelines, whether model-based or data-driven, often constrained by complex and fragmented workflow, rely heavily on expert knowledge and manual engineering, and struggle with adaptability and generalization under limited data. In contrast, Large Language Models (LLMs) offer strong reasoning capabilities, broad general-purpose knowledge, in-context learning, and cross-modal transfer abilities, positioning them as powerful tools for automating and generalizing SP workflows. Motivated by these potentials, we introduce SignalLLM, the first general-purpose LLM-based agent framework for general SP tasks. Unlike prior LLM-based SP approaches that are limited to narrow applications or tricky prompting, SignalLLM introduces a principled, modular architecture. It decomposes high-level SP goals into structured subtasks via in-context learning and domain-specific retrieval, followed by hierarchical planning through adaptive retrieval-augmented generation (RAG) and refinement; these subtasks are then executed through prompt-based reasoning, cross-modal reasoning, code synthesis, model invocation, or data-driven LLM-assisted modeling. Its generalizable design enables the flexible selection of problem solving strategies across different signal modalities, task types, and data conditions. We demonstrate the versatility and effectiveness of SignalLLM through five representative tasks in communication and sensing, such as radar target detection, human activity recognition, and text compression. Experimental results show superior performance over traditional and existing LLM-based methods, particularly in few-shot and zero-shot settings.

comment: 11 pages

♻ ☆ AI's Social Forcefield: Reshaping Distributed Cognition in Human-AI Teams

Christoph Riedl, Saiph Savage, Josie Zvelebilova

AI is not only a neutral tool in team settings; it actively reshapes the social and cognitive fabric of collaboration. We advance a unified framework of alignment in distributed cognition in human-AI teams -- a process through which linguistic, cognitive, and social coordination emerge as human and AI agents co-construct a shared representational space. Across two studies, we show that exposure to AI-generated language shapes not only how people speak, but also how they think, what they attend to, and how they relate to each other. Together, these findings reveal how AI participation reorganizes the distributed cognitive architecture of teams: AI systems function as implicit social forcefields. Our findings highlight the double-edged impact of AI: the same mechanisms that enable efficient collaboration can also erode epistemic diversity and undermine natural alignment processes. We argue for rethinking AI in teams as a socially influential actor and call for new design paradigms that foreground transparency, controllability, and group-level dynamics to foster responsible, productive human-AI collaboration.

♻ ☆ Epistemic Diversity and Knowledge Collapse in Large Language Models

Dustin Wright, Sarah Masud, Jared Moore, Srishti Yadav, Maria Antoniak, Chan Young Park, Isabelle Augenstein

Large language models (LLMs) tend to generate lexically, semantically, and stylistically homogenous texts. This poses a risk of knowledge collapse, where homogenous LLMs mediate a shrinking in the range of accessible information over time. Existing works on homogenization are limited by a focus on closed-ended multiple-choice setups or fuzzy semantic features, and do not look at trends across time and cultural contexts. To overcome this, we present a new methodology to measure epistemic diversity, i.e., variation in real-world claims in LLM outputs, which we use to perform a broad empirical study of LLM knowledge collapse. We test 27 LLMs, 155 topics covering 12 countries, and 200 prompt variations sourced from real user chats. For the topics in our study, we show that while newer models tend to generate more diverse claims, nearly all models are less epistemically diverse than a basic web search. We find that model size has a negative impact on epistemic diversity, while retrieval-augmented generation (RAG) has a positive impact, though the improvement from RAG varies by the cultural context. Finally, compared to a traditional knowledge source (Wikipedia), we find that country-specific claims reflect the English language more than the local one, highlighting a gap in epistemic representation

comment: 16 pages; 8 figures, 4 tables; v2 changelog: Fixed the modeling for table 3, random effect is the model version; v3 changelog: Fixed minor formatting issues in tables 2 and 3; v4 changelog: Fixed some typos and model description

♻ ☆ Incentivizing LLMs to Self-Verify Their Answers

Fuxiang Zhang, Jiacheng Xu, Chaojie Wang, Ce Cui, Yang Liu, Bo An

Large Language Models (LLMs) have demonstrated remarkable progress in complex reasoning tasks through both post-training and test-time scaling laws. While prevalent test-time scaling approaches are often realized by using external reward models to guide the model generation process, we find that only marginal gains can be acquired when scaling a model post-trained on specific reasoning tasks. We identify that the limited improvement stems from distribution discrepancies between the specific post-trained generator and the general reward model. To address this, we propose a framework that incentivizes LLMs to self-verify their own answers. By unifying answer generation and verification within a single reinforcement learning (RL) process, we train models that can effectively assess the correctness of their own solutions. The trained model can further scale its performance at inference time by verifying its generations, without the need for external verifiers. We train our self-verification models based on Qwen2.5-Math-7B and DeepSeek-R1-Distill-Qwen-1.5B, demonstrating their capabilities across varying reasoning context lengths. Experiments on multiple mathematical reasoning benchmarks show that our models can not only improve post-training performance but also enable effective test-time scaling.

♻ ☆ Human-Like Goalkeeping in a Realistic Football Simulation: a Sample-Efficient Reinforcement Learning Approach

Alessandro Sestini, Joakim Bergdahl, Jean-Philippe Barrette-LaPierre, Florian Fuchs, Brady Chen, Michael Jones, Linus Gisslén

While several high profile video games have served as testbeds for Deep Reinforcement Learning (DRL), this technique has rarely been employed by the game industry for crafting authentic AI behaviors. Previous research focuses on training super-human agents with large models, which is impractical for game studios with limited resources aiming for human-like agents. This paper proposes a sample-efficient DRL method tailored for training and fine-tuning agents in industrial settings such as the video game industry. Our method improves sample efficiency of value-based DRL by leveraging pre-collected data and increasing network plasticity. We evaluate our method training a goalkeeper agent in EA SPORTS FC 25, one of the best-selling football simulations today. Our agent outperforms the game's built-in AI by 10% in ball saving rate. Ablation studies show that our method trains agents 50% faster compared to standard DRL methods. Finally, qualitative evaluation from domain experts indicates that our approach creates more human-like gameplay compared to hand-crafted agents. As a testament to the impact of the approach, the method has been adopted for use in the most recent release of the series.

♻ ☆ A mathematical certification for positivity conditions in Neural Networks with applications to partial monotonicity and Trustworthy AI

Alejandro Polo-Molina, David Alfaya, Jose Portela

Artificial Neural Networks (ANNs) have become a powerful tool for modeling complex relationships in large-scale datasets. However, their black-box nature poses trustworthiness challenges. In certain situations, ensuring trust in predictions might require following specific partial monotonicity constraints. However, certifying if an already-trained ANN is partially monotonic is challenging. Therefore, ANNs are often disregarded in some critical applications, such as credit scoring, where partial monotonicity is required. To address this challenge, this paper presents a novel algorithm (LipVor) that certifies if a black-box model, such as an ANN, is positive based on a finite number of evaluations. Consequently, since partial monotonicity can be expressed as a positivity condition on partial derivatives, LipVor can certify whether an ANN is partially monotonic. To do so, for every positively evaluated point, the Lipschitzianity of the black-box model is used to construct a specific neighborhood where the function remains positive. Next, based on the Voronoi diagram of the evaluated points, a sufficient condition is stated to certify if the function is positive in the domain. Unlike prior methods, our approach certifies partial monotonicity without constrained architectures or piece-wise linear activations. Therefore, LipVor could open up the possibility of using unconstrained ANN in some critical fields. Moreover, some other properties of an ANN, such as convexity, can be posed as positivity conditions, and therefore, LipVor could also be applied.

comment: 16 pages, 4 figures

♻ ☆ Collab-REC: An LLM-based Agentic Framework for Balancing Recommendations in Tourism

Ashmi Banerjee, Adithi Satish, Fitri Nur Aisyah, Wolfgang Wörndl, Yashar Deldjoo

We propose Collab-REC, a multi-agent framework designed to counteract popularity bias and enhance diversity in tourism recommendations. In our setting, three LLM-based agents -- Personalization, Popularity, and Sustainability generate city suggestions from complementary perspectives. A non-LLM moderator then merges and refines these proposals via multi-round negotiation, ensuring each agent's viewpoint is incorporated while penalizing spurious or repeated responses. Experiments on European city queries show that Collab-REC improves diversity and overall relevance compared to a single-agent baseline, surfacing lesser-visited locales that often remain overlooked. This balanced, context-aware approach addresses over-tourism and better aligns with constraints provided by the user, highlighting the promise of multi-stakeholder collaboration in LLM-driven recommender systems.

♻ ☆ UV-Attack: Physical-World Adversarial Attacks for Person Detection via Dynamic-NeRF-based UV Mapping ICLR2025

Yanjie Li, Kaisheng Liang, Bin Xiao

In recent research, adversarial attacks on person detectors using patches or static 3D model-based texture modifications have struggled with low success rates due to the flexible nature of human movement. Modeling the 3D deformations caused by various actions has been a major challenge. Fortunately, advancements in Neural Radiance Fields (NeRF) for dynamic human modeling offer new possibilities. In this paper, we introduce UV-Attack, a groundbreaking approach that achieves high success rates even with extensive and unseen human actions. We address the challenge above by leveraging dynamic-NeRF-based UV mapping. UV-Attack can generate human images across diverse actions and viewpoints, and even create novel actions by sampling from the SMPL parameter space. While dynamic NeRF models are capable of modeling human bodies, modifying clothing textures is challenging because they are embedded in neural network parameters. To tackle this, UV-Attack generates UV maps instead of RGB images and modifies the texture stacks. This approach enables real-time texture edits and makes the attack more practical. We also propose a novel Expectation over Pose Transformation loss (EoPT) to improve the evasion success rate on unseen poses and views. Our experiments show that UV-Attack achieves a 92.7% attack success rate against the FastRCNN model across varied poses in dynamic video settings, significantly outperforming the state-of-the-art AdvCamou attack, which only had a 28.5% ASR. Moreover, we achieve 49.5% ASR on the latest YOLOv8 detector in black-box settings. This work highlights the potential of dynamic NeRF-based UV mapping for creating more effective adversarial attacks on person detectors, addressing key challenges in modeling human movement and texture modification. The code is available at https://github.com/PolyLiYJ/UV-Attack.

comment: 23 pages, 22 figures, accepted by ICLR2025

♻ ☆ Rethinking Optimal Verification Granularity for Compute-Efficient Test-Time Scaling NeurIPS 2025

Hao Mark Chen, Guanxi Lu, Yasuyuki Okoshi, Zhiwen Mo, Masato Motomura, Hongxiang Fan

Test-time scaling (TTS) has proven effective in enhancing the reasoning capabilities of large language models (LLMs). Verification plays a key role in TTS, simultaneously influencing (1) reasoning performance and (2) compute efficiency, due to the quality and computational cost of verification. In this work, we challenge the conventional paradigms of verification, and make the first attempt toward systematically investigating the impact of verification granularity-that is, how frequently the verifier is invoked during generation, beyond verifying only the final output or individual generation steps. To this end, we introduce Variable Granularity Search (VG-Search), a unified algorithm that generalizes beam search and Best-of-N sampling via a tunable granularity parameter g. Extensive experiments with VG-Search under varying compute budgets, generator-verifier configurations, and task attributes reveal that dynamically selecting g can improve the compute efficiency and scaling behavior. Building on these findings, we propose adaptive VG-Search strategies that achieve accuracy gains of up to 3.1\% over Beam Search and 3.6\% over Best-of-N, while reducing FLOPs by over 52\%. We will open-source the code to support future research.

comment: Accepted at NeurIPS 2025

♻ ☆ More of the Same: Persistent Representational Harms Under Increased Representation NeurIPS 2025

Jennifer Mickel, Maria De-Arteaga, Leqi Liu, Kevin Tian

To recognize and mitigate the harms of generative AI systems, it is crucial to consider who is represented in the outputs of generative AI systems and how people are represented. A critical gap emerges when naively improving who is represented, as this does not imply bias mitigation efforts have been applied to address how people are represented. We critically examined this by investigating gender representation in occupation across state-of-the-art large language models. We first show evidence suggesting that over time there have been interventions to models altering the resulting gender distribution, and we find that women are more represented than men when models are prompted to generate biographies or personas. We then demonstrate that representational biases persist in how different genders are represented by examining statistically significant word differences across genders. This results in a proliferation of representational harms, stereotypes, and neoliberalism ideals that, despite existing interventions to increase female representation, reinforce existing systems of oppression.

comment: Accepted by the 39th Conference on Neural Information Processing Systems (NeurIPS 2025) as a poster paper; 39 pages, 7 figures, 15 tables

♻ ☆ VerifIoU -- Robustness of Object Detection to Perturbations SC

Noémie Cohen, Mélanie Ducoffe, Ryma Boumazouza, Christophe Gabreau, Claire Pagetti, Xavier Pucel, Audrey Galametz

We introduce a novel Interval Bound Propagation (IBP) approach for the formal verification of object detection models, specifically targeting the Intersection over Union (IoU) metric. The approach has been implemented in an open source code, named IBP IoU, compatible with popular abstract interpretation based verification tools. The resulting verifier is evaluated on landing approach runway detection and handwritten digit recognition case studies. Comparisons against a baseline (Vanilla IBP IoU) highlight the superior performance of IBP IoU in ensuring accuracy and stability, contributing to more secure and robust machine learning applications.

comment: 44th Digital Avionics Systems Conference (DASC), Sep 2025, Montreal, Canada

♻ ☆ MedAgentBoard: Benchmarking Multi-Agent Collaboration with Conventional Methods for Diverse Medical Tasks NeurIPS 2025

Yinghao Zhu, Ziyi He, Haoran Hu, Xiaochen Zheng, Xichen Zhang, Zixiang Wang, Junyi Gao, Liantao Ma, Lequan Yu

The rapid advancement of Large Language Models (LLMs) has stimulated interest in multi-agent collaboration for addressing complex medical tasks. However, the practical advantages of multi-agent collaboration approaches remain insufficiently understood. Existing evaluations often lack generalizability, failing to cover diverse tasks reflective of real-world clinical practice, and frequently omit rigorous comparisons against both single-LLM-based and established conventional methods. To address this critical gap, we introduce MedAgentBoard, a comprehensive benchmark for the systematic evaluation of multi-agent collaboration, single-LLM, and conventional approaches. MedAgentBoard encompasses four diverse medical task categories: (1) medical (visual) question answering, (2) lay summary generation, (3) structured Electronic Health Record (EHR) predictive modeling, and (4) clinical workflow automation, across text, medical images, and structured EHR data. Our extensive experiments reveal a nuanced landscape: while multi-agent collaboration demonstrates benefits in specific scenarios, such as enhancing task completeness in clinical workflow automation, it does not consistently outperform advanced single LLMs (e.g., in textual medical QA) or, critically, specialized conventional methods that generally maintain better performance in tasks like medical VQA and EHR-based prediction. MedAgentBoard offers a vital resource and actionable insights, emphasizing the necessity of a task-specific, evidence-based approach to selecting and developing AI solutions in medicine. It underscores that the inherent complexity and overhead of multi-agent collaboration must be carefully weighed against tangible performance gains. All code, datasets, detailed prompts, and experimental results are open-sourced at https://medagentboard.netlify.app/.

comment: Accepted by NeurIPS 2025 Datasets & Benchmarks Track

♻ ☆ Demystifying the Roles of LLM Layers in Retrieval, Knowledge, and Reasoning ICASSP 2025

Xinyuan Song, Keyu Wang, PengXiang Li, Lu Yin, Shiwei Liu

Recent studies suggest that the deeper layers of Large Language Models (LLMs) contribute little to representation learning and can often be removed without significant performance loss. However, such claims are typically drawn from narrow evaluations and may overlook important aspects of model behavior. In this work, we present a systematic study of depth utilization across diverse dimensions, including evaluation protocols, task categories, and model architectures. Our analysis confirms that very deep layers are generally less effective than earlier ones, but their contributions vary substantially with the evaluation setting. Under likelihood-based metrics without generation, pruning most layers preserves performance, with only the initial few being critical. By contrast, generation-based evaluation uncovers indispensable roles for middle and deeper layers in enabling reasoning and maintaining long-range coherence. We further find that knowledge and retrieval are concentrated in shallow components, whereas reasoning accuracy relies heavily on deeper layers -- yet can be reshaped through distillation. These results highlight that depth usage in LLMs is highly heterogeneous and context-dependent, underscoring the need for task-, metric-, and model-aware perspectives in both interpreting and compressing large models.

comment: ICASSP 2025

♻ ☆ StyleGuard: Preventing Text-to-Image-Model-based Style Mimicry Attacks by Style Perturbations NIPS2025

Yanjie Li, Wenxuan Zhang, Xinqi Lyu, Yihao Liu, Bin Xiao

Recently, text-to-image diffusion models have been widely used for style mimicry and personalized customization through methods such as DreamBooth and Textual Inversion. This has raised concerns about intellectual property protection and the generation of deceptive content. Recent studies, such as Glaze and Anti-DreamBooth, have proposed using adversarial noise to protect images from these attacks. However, recent purification-based methods, such as DiffPure and Noise Upscaling, have successfully attacked these latest defenses, showing the vulnerabilities of these methods. Moreover, present methods show limited transferability across models, making them less effective against unknown text-to-image models. To address these issues, we propose a novel anti-mimicry method, StyleGuard. We propose a novel style loss that optimizes the style-related features in the latent space to make it deviate from the original image, which improves model-agnostic transferability. Additionally, to enhance the perturbation's ability to bypass diffusion-based purification, we designed a novel upscale loss that involves ensemble purifiers and upscalers during training. Extensive experiments on the WikiArt and CelebA datasets demonstrate that StyleGuard outperforms existing methods in robustness against various transformations and purifications, effectively countering style mimicry in various models. Moreover, StyleGuard is effective on different style mimicry methods, including DreamBooth and Textual Inversion. The code is available at https://github.com/PolyLiYJ/StyleGuard.

comment: Accepted by NIPS2025

♻ ☆ Toward a Public and Secure Generative AI: A Comparative Analysis of Open and Closed LLMs

Jorge Machado

Generative artificial intelligence (Gen AI) systems represent a critical technology with far-reaching implications across multiple domains of society. However, their deployment entails a range of risks and challenges that require careful evaluation. To date, there has been a lack of comprehensive, interdisciplinary studies offering a systematic comparison between open-source and proprietary (closed) generative AI systems, particularly regarding their respective advantages and drawbacks. This study aims to: i) critically evaluate and compare the characteristics, opportunities, and challenges of open and closed generative AI models; and ii) propose foundational elements for the development of an Open, Public, and Safe Gen AI framework. As a methodology, we adopted a combined approach that integrates three methods: literature review, critical analysis, and comparative analysis. The proposed framework outlines key dimensions, openness, public governance, and security, as essential pillars for shaping the future of trustworthy and inclusive Gen AI. Our findings reveal that open models offer greater transparency, auditability, and flexibility, enabling independent scrutiny and bias mitigation. In contrast, closed systems often provide better technical support and ease of implementation, but at the cost of unequal access, accountability, and ethical oversight. The research also highlights the importance of multi-stakeholder governance, environmental sustainability, and regulatory frameworks in ensuring responsible development.

♻ ☆ Monopoly Deal: A Benchmark Environment for Bounded One-Sided Response Games

Will Wolf

Card games are widely used to study sequential decision-making under uncertainty, with real-world analogues in negotiation, finance, and cybersecurity. These games typically fall into three categories based on the flow of control: strictly sequential (players alternate single actions), deterministic response (some actions trigger a fixed outcome), and unbounded reciprocal response (alternating counterplays are permitted). A less-explored but strategically rich structure is the bounded one-sided response, where a player's action briefly transfers control to the opponent, who must satisfy a fixed condition through one or more moves before the turn resolves. We term games featuring this mechanism Bounded One-Sided Response Games (BORGs). We introduce a modified version of Monopoly Deal as a benchmark environment that isolates this dynamic, where a Rent action forces the opponent to choose payment assets. The gold-standard algorithm, Counterfactual Regret Minimization (CFR), converges on effective strategies without novel algorithmic extensions. A lightweight full-stack research platform unifies the environment, a parallelized CFR runtime, and a human-playable web interface. The trained CFR agent and source code are available at https://monopolydeal.ai.

comment: 24 pages, 7 figures

♻ ☆ Lost in Tokenization: Context as the Key to Unlocking Biomolecular Understanding in Scientific LLMs

Kai Zhuang, Jiawei Zhang, Yumou Liu, Hanqun Cao, Chunbin Gu, Mengdi Liu, Zhangyang Gao, Zitong Jerry Wang, Xuanhe Zhou, Pheng-Ann Heng, Lijun Wu, Conghui He, Cheng Tan

Scientific Large Language Models (Sci-LLMs) have emerged as a promising frontier for accelerating biological discovery. However, these models face a fundamental challenge when processing raw biomolecular sequences: the tokenization dilemma. Whether treating sequences as a specialized language, risking the loss of functional motif information, or as a separate modality, introducing formidable alignment challenges, current strategies fundamentally limit their reasoning capacity. We challenge this sequence-centric paradigm by positing that a more effective strategy is to provide Sci-LLMs with high-level structured context derived from established bioinformatics tools, thereby bypassing the need to interpret low-level noisy sequence data directly. Through a systematic comparison of leading Sci-LLMs on biological reasoning tasks, we tested three input modes: sequence-only, context-only, and a combination of both. Our findings are striking: the context-only approach consistently and substantially outperforms all other modes. Even more revealing, the inclusion of the raw sequence alongside its high-level context consistently degrades performance, indicating that raw sequences act as informational noise, even for models with specialized tokenization schemes. These results suggest that the primary strength of existing Sci-LLMs lies not in their nascent ability to interpret biomolecular syntax from scratch, but in their profound capacity for reasoning over structured, human-readable knowledge. Therefore, we argue for reframing Sci-LLMs not as sequence decoders, but as powerful reasoning engines over expert knowledge. This work lays the foundation for a new class of hybrid scientific AI agents, repositioning the developmental focus from direct sequence interpretation towards high-level knowledge synthesis. The code is available at https://github.com/opendatalab-raiser/CoKE.

comment: 38 pages, under review

♻ ☆ In Defence of Post-hoc Explainability NeurIPS 2024

Nick Oh

This position paper defends post-hoc explainability methods as legitimate tools for scientific knowledge production in machine learning. Addressing criticism of these methods' reliability and epistemic status, we develop a philosophical framework grounded in mediated understanding and bounded factivity. We argue that scientific insights can emerge through structured interpretation of model behaviour without requiring complete mechanistic transparency, provided explanations acknowledge their approximative nature and undergo rigorous empirical validation. Through analysis of recent biomedical ML applications, we demonstrate how post-hoc methods, when properly integrated into scientific practice, generate novel hypotheses and advance phenomenal understanding.

comment: v1 presented at the Interpretable AI: Past, Present, and Future Workshop at NeurIPS 2024 (non-archival)

♻ ☆ On-the-Fly OVD Adaptation with FLAME: Few-shot Localization via Active Marginal-Samples Exploration

Yehonathan Refael, Amit Aides, Aviad Barzilai, George Leifman, Genady Beryozkin, Vered Silverman, Bolous Jaber, Tomer Shekel

Open-vocabulary object detection (OVD) models offer remarkable flexibility by detecting objects from arbitrary text queries. However, their zero-shot performance in specialized domains like Remote Sensing (RS) is often compromised by the inherent ambiguity of natural language, limiting critical downstream applications. For instance, an OVD model may struggle to distinguish between fine-grained classes such as "fishing boat" and "yacht" since their embeddings are similar and often inseparable. This can hamper specific user goals, such as monitoring illegal fishing, by producing irrelevant detections. To address this, we propose a cascaded approach that couples the broad generalization of a large pre-trained OVD model with a lightweight few-shot classifier. Our method first employs the zero-shot model to generate high-recall object proposals. These proposals are then refined for high precision by a compact classifier trained in real-time on only a handful of user-annotated examples - drastically reducing the high costs of RS imagery annotation.The core of our framework is FLAME, a one-step active learning strategy that selects the most informative samples for training. FLAME identifies, on the fly, uncertain marginal candidates near the decision boundary using density estimation, followed by clustering to ensure sample diversity. This efficient sampling technique achieves high accuracy without costly full-model fine-tuning and enables instant adaptation, within less then a minute, which is significantly faster than state-of-the-art alternatives.Our method consistently surpasses state-of-the-art performance on RS benchmarks, establishing a practical and resource-efficient framework for adapting foundation models to specific user needs.

♻ ☆ Tunable-Generalization Diffusion Powered by Self-Supervised Contextual Sub-Data for Low-Dose CT Reconstruction

Guoquan Wei, Liu Shi, Zekun Zhou, Wenzhe Shan, Qiegen Liu

Current models based on deep learning for low-dose CT denoising rely heavily on paired data and generalize poorly. Even the more concerned diffusion models need to learn the distribution of clean data for reconstruction, which is difficult to satisfy in medical clinical applications. At the same time, self-supervised-based methods face the challenge of significant degradation of generalizability of models pre-trained for the current dose to expand to other doses. To address these issues, this work proposes a novel method of TUnable-geneRalizatioN Diffusion (TurnDiff) powered by self-supervised contextual sub-data for low-dose CT reconstruction. Firstly, a contextual subdata self-enhancing similarity strategy is designed for denoising centered on the LDCT projection domain, which provides an initial prior for the subsequent progress. Subsequently, the initial prior is used to combine knowledge distillation with a deep combination of latent diffusion models for optimizing image details. The pre-trained model is used for inference reconstruction, and the pixel-level self-correcting fusion technique is proposed for fine-grained reconstruction of the image domain to enhance the image fidelity, using the initial prior and the LDCT image as a guide. In addition, the technique is flexibly applied to the generalization of upper and lower doses or even unseen doses. Dual-domain strategy cascade for self-supervised LDCT denoising, TurnDiff requires only LDCT projection domain data for training and testing. Comprehensive evaluation on both benchmark datasets and real-world data demonstrates that TurnDiff consistently outperforms state-of-the-art methods in both reconstruction and generalization.

♻ ☆ Towards a Method for Synthetic Generation of Persons with Aphasia Transcripts

Jason M. Pittman, Anton Phillips Jr., Yesenia Medina-Santos, Brielle C. Stark

In aphasia research, Speech-Language Pathologists (SLPs) devote extensive time to manually coding speech samples using Correct Information Units (CIUs), a measure of how informative an individual sample of speech is. Developing automated systems to recognize aphasic language is limited by data scarcity. For example, only about 600 transcripts are available in AphasiaBank yet billions of tokens are used to train large language models (LLMs). In the broader field of machine learning (ML), researchers increasingly turn to synthetic data when such are sparse. Therefore, this study constructs and validates two methods to generate synthetic transcripts of the AphasiaBank Cat Rescue picture description task. One method leverages a procedural programming approach while the second uses Mistral 7b Instruct and Llama 3.1 8b Instruct LLMs. The methods generate transcripts across four severity levels (Mild, Moderate, Severe, Very Severe) through word dropping, filler insertion, and paraphasia substitution. Overall, we found, compared to human-elicited transcripts, Mistral 7b Instruct best captures key aspects of linguistic degradation observed in aphasia, showing realistic directional changes in NDW, word count, and word length amongst the synthetic generation methods. Based on the results, future work should plan to create a larger dataset, fine-tune models for better aphasic representation, and have SLPs assess the realism and usefulness of the synthetic transcripts.

comment: 19 pages, 1 figure, 7 tables

♻ ☆ Paper2Poster: Towards Multimodal Poster Automation from Scientific Papers

Wei Pang, Kevin Qinghong Lin, Xiangru Jian, Xi He, Philip Torr

Academic poster generation is a crucial yet challenging task in scientific communication, requiring the compression of long-context interleaved documents into a single, visually coherent page. To address this challenge, we introduce the first benchmark and metric suite for poster generation, which pairs recent conference papers with author-designed posters and evaluates outputs on (i)Visual Quality-semantic alignment with human posters, (ii)Textual Coherence-language fluency, (iii)Holistic Assessment-six fine-grained aesthetic and informational criteria scored by a VLM-as-judge, and notably (iv)PaperQuiz-the poster's ability to convey core paper content as measured by VLMs answering generated quizzes. Building on this benchmark, we propose PosterAgent, a top-down, visual-in-the-loop multi-agent pipeline: the (a)Parser distills the paper into a structured asset library; the (b)Planner aligns text-visual pairs into a binary-tree layout that preserves reading order and spatial balance; and the (c)Painter-Commenter loop refines each panel by executing rendering code and using VLM feedback to eliminate overflow and ensure alignment. In our comprehensive evaluation, we find that GPT-4o outputs-though visually appealing at first glance-often exhibit noisy text and poor PaperQuiz scores, and we find that reader engagement is the primary aesthetic bottleneck, as human-designed posters rely largely on visual semantics to convey meaning. Our fully open-source variants (e.g. based on the Qwen-2.5 series) outperform existing 4o-driven multi-agent systems across nearly all metrics, while using 87% fewer tokens. It transforms a 22-page paper into a finalized yet editable .pptx poster - all for just $0.005. These findings chart clear directions for the next generation of fully automated poster-generation models. The code and datasets are available at https://github.com/Paper2Poster/Paper2Poster.

comment: Project Page: https://github.com/Paper2Poster/Paper2Poster

♻ ☆ BhashaBench V1: A Comprehensive Benchmark for the Quadrant of Indic Domains

Vijay Devane, Mohd Nauman, Bhargav Patel, Aniket Mahendra Wakchoure, Yogeshkumar Sant, Shyam Pawar, Viraj Thakur, Ananya Godse, Sunil Patra, Neha Maurya, Suraj Racha, Nitish Kamal Singh, Ajay Nagpal, Piyush Sawarkar, Kundeshwar Vijayrao Pundalik, Rohit Saluja, Ganesh Ramakrishnan

The rapid advancement of large language models(LLMs) has intensified the need for domain and culture specific evaluation. Existing benchmarks are largely Anglocentric and domain-agnostic, limiting their applicability to India-centric contexts. To address this gap, we introduce BhashaBench V1, the first domain-specific, multi-task, bilingual benchmark focusing on critical Indic knowledge systems. BhashaBench V1 contains 74,166 meticulously curated question-answer pairs, with 52,494 in English and 21,672 in Hindi, sourced from authentic government and domain-specific exams. It spans four major domains: Agriculture, Legal, Finance, and Ayurveda, comprising 90+ subdomains and covering 500+ topics, enabling fine-grained evaluation. Evaluation of 29+ LLMs reveals significant domain and language specific performance gaps, with especially large disparities in low-resource domains. For instance, GPT-4o achieves 76.49% overall accuracy in Legal but only 59.74% in Ayurveda. Models consistently perform better on English content compared to Hindi across all domains. Subdomain-level analysis shows that areas such as Cyber Law, International Finance perform relatively well, while Panchakarma, Seed Science, and Human Rights remain notably weak. BhashaBench V1 provides a comprehensive dataset for evaluating large language models across India's diverse knowledge domains. It enables assessment of models' ability to integrate domain-specific knowledge with bilingual understanding. All code, benchmarks, and resources are publicly available to support open research.

♻ ☆ MindGYM: What Matters in Question Synthesis for Thinking-Centric Fine-Tuning? NeurIPS'25

Zhe Xu, Daoyuan Chen, Zhenqing Ling, Yaliang Li, Ying Shen

Large foundation models face challenges in acquiring transferable, structured thinking abilities, especially when supervised with rigid templates or crowd-annotated instruction datasets. Unlike prior approaches, we focus on a thinking-centric data synthesis paradigm that enables models to evolve through self-generated, cognitively guided data. We propose MindGYM, a structured and scalable framework for question synthesis, composed of: (1) Cognitive Thinking Process Injection, which infuses high-level reasoning objectives to shape the model's synthesis behavior; (2) Seed Single-Hop Question Synthesis, generating atomic questions from diverse semantic types to encourage broader thinking; and (3) Challenging Multi-Hop QA Synthesis, composing more complex multi-hop questions based on QA seeds for deeper reasoning. Detailed analysis shows that synthetic data generated by our method achieves 16.7% higher average quality and 67.91% lower quality variance compared to baseline sources, highlighting that both high-quality and self-contained data are essential for effective, thinking-oriented fine-tuning. MindGYM improves performance on six reasoning benchmarks, achieving gains of up to 16% on MathVision using only 400 data samples, and generalizable improvements across different model sizes and architectures. MindGYM underscores the viability of self-challenging mechanisms in refining large model capabilities while minimizing human intervention and resource demands. Code and data are released to promote data-centric research into self-evolving foundation models driven by their internal reasoning capabilities.

comment: Accepted by NeurIPS'25. 30 pages, 2 figures, 13 tables

♻ ☆ AIMeter: Measuring, Analyzing, and Visualizing Energy and Carbon Footprint of AI Workloads

Hongzhen Huang, Kunming Zhang, Hanlong Liao, Kui Wu, Guoming Tang

The rapid advancement of AI, particularly large language models (LLMs), has raised significant concerns about the energy use and carbon emissions associated with model training and inference. However, existing tools for measuring and reporting such impacts are often fragmented, lacking systematic metric integration and offering limited support for correlation analysis among them. This paper presents AIMeter, a comprehensive software toolkit for the measurement, analysis, and visualization of energy use, power draw, hardware performance, and carbon emissions across AI workloads. By seamlessly integrating with existing AI frameworks, AIMeter offers standardized reports and exports fine-grained time-series data to support benchmarking and reproducibility in a lightweight manner. It further enables in-depth correlation analysis between hardware metrics and model performance and thus facilitates bottleneck identification and performance enhancement. By addressing critical limitations in existing tools, AIMeter encourages the research community to weigh environmental impact alongside raw performance of AI workloads and advances the shift toward more sustainable "Green AI" practices. The code is available at https://github.com/SusCom-Lab/AIMeter.

comment: 11 pages, 7 figures and 5 tables

♻ ☆ UNO-Bench: A Unified Benchmark for Exploring the Compositional Law Between Uni-modal and Omni-modal in Omni Models

Chen Chen, ZeYang Hu, Fengjiao Chen, Liya Ma, Jiaxing Liu, Xiaoyu Li, Ziwen Wang, Xuezhi Cao, Xunliang Cai

Multimodal Large Languages models have been progressing from uni-modal understanding toward unifying visual, audio and language modalities, collectively termed omni models. However, the correlation between uni-modal and omni-modal remains unclear, which requires comprehensive evaluation to drive omni model's intelligence evolution. In this work, we introduce a novel, high-quality, and UNified Omni model benchmark, UNO-Bench. This benchmark is designed to effectively evaluate both UNi-modal and Omni-modal capabilities under a unified ability taxonomy, spanning 44 task types and 5 modality combinations. It includes 1250 human curated samples for omni-modal with 98% cross-modality solvability, and 2480 enhanced uni-modal samples. The human-generated dataset is well-suited to real-world scenarios, particularly within the Chinese context, whereas the automatically compressed dataset offers a 90% increase in speed and maintains 98% consistency across 18 public benchmarks. In addition to traditional multi-choice questions, we propose an innovative multi-step open-ended question format to assess complex reasoning. A general scoring model is incorporated, supporting 6 question types for automated evaluation with 95% accuracy. Experimental result shows the Compositional Law between omni-modal and uni-modal performance and the omni-modal capability manifests as a bottleneck effect on weak models, while exhibiting synergistic promotion on strong models.

comment: v3: Switch the paper template. Work in progress. Github: https://github.com/meituan-longcat/UNO-Bench Hugging Face: https://huggingface.co/datasets/meituan-longcat/UNO-Bench

♻ ☆ Reflection on Data Storytelling Tools in the Generative AI Era from the Human-AI Collaboration Perspective IEEE VIS 25

Haotian Li, Yun Wang, Huamin Qu

Human-AI collaborative tools attract attentions from the data storytelling community to lower the expertise barrier and streamline the workflow. The recent advance in large-scale generative AI techniques, e.g., large language models (LLMs) and text-to-image models, has the potential to enhance data storytelling with their power in visual and narration generation. After two years since these techniques were publicly available, it is important to reflect our progress of applying them and have an outlook for future opportunities. To achieve the goal, we compare the collaboration patterns of the latest tools with those of earlier ones using a dedicated framework for understanding human-AI collaboration in data storytelling. Through comparison, we identify consistently widely studied patterns, e.g., human-creator + AI-assistant, and newly explored or emerging ones, e.g., AI-creator + human-reviewer. The benefits of these AI techniques and implications to human-AI collaboration are also revealed. We further propose future directions to hopefully ignite innovations.

comment: This paper is a sequel to the CHI 24 paper "Where Are We So Far? Understanding Data Storytelling Tools from the Perspective of Human-AI Collaboration (https://doi.org/10.1145/3613904.3642726), aiming to refresh our understanding with the latest advancements. It is accepted at IEEE VIS 25

♻ ☆ Diversity as a Reward: Fine-Tuning LLMs on a Mixture of Domain-Undetermined Data NeurIPS'25

Zhenqing Ling, Daoyuan Chen, Liuyi Yao, Qianli Shen, Yaliang Li, Ying Shen

Fine-tuning large language models (LLMs) using diverse datasets is crucial for enhancing their overall performance across various domains. In practical scenarios, existing methods based on modeling the mixture proportions of data composition often struggle with data whose domain labels are missing, imprecise or non-normalized, while methods based on data selection usually encounter difficulties in balancing multi-domain performance. To address these challenges, in this work, we investigate the role of data diversity in enhancing the overall abilities of LLMs by empirically constructing contrastive data pools and theoretically deriving explanations. Building upon the insights gained, we propose a new method that gives the LLM a dual identity: an output model to cognitively probe and select data based on diversity reward, as well as an input model to be tuned with the selected data. Extensive experiments show that the proposed method notably boosts performance across domain-undetermined data and a series of foundational downstream tasks when applied to various advanced LLMs. We release our code and hope this study can shed light on the understanding of data diversity and advance feedback-driven data-model co-design for LLMs.

comment: Accepted by NeurIPS'25 main track. 47 pages, 21 figures, 32 tables

♻ ☆ Model-Document Protocol for AI Search

Hongjin Qian, Zheng Liu

AI search depends on linking large language models (LLMs) with vast external knowledge sources. Yet web pages, PDF files, and other raw documents are not inherently LLM-ready: they are long, noisy, and unstructured. Conventional retrieval methods treat these documents as verbatim text and return raw passages, leaving the burden of fragment assembly and contextual reasoning to the LLM. This gap underscores the need for a new retrieval paradigm that redefines how models interact with documents. We introduce the Model-Document Protocol (MDP), a general framework that formalizes how raw text is bridged to LLMs through consumable knowledge representations. Rather than treating retrieval as passage fetching, MDP defines multiple pathways that transform unstructured documents into task-specific, LLM-ready inputs. These include agentic reasoning, which curates raw evidence into coherent context; memory grounding, which accumulates reusable notes to enrich reasoning; and structured leveraging, which encodes documents into formal representations such as graphs or key-value caches. All three pathways share the same goal: ensuring that what reaches the LLM is not raw fragments but compact, structured knowledge directly consumable for reasoning. As an instantiation, we present MDP-Agent, which realizes the protocol through an agentic process: constructing document-level gist memories for global coverage, performing diffusion-based exploration with vertical exploitation to uncover layered dependencies, and applying map-reduce style synthesis to integrate large-scale evidence into compact yet sufficient context. Experiments on information-seeking benchmarks demonstrate that MDP-Agent outperforms baselines, validating both the soundness of the MDP framework and the effectiveness of its agentic instantiation.

comment: 10 pages

♻ ☆ Chaos-based reinforcement learning with TD3

Toshitaka Matsuki, Yusuke Sakemi, Kazuyuki Aihara

Chaos-based reinforcement learning (CBRL) is a method in which the agent's internal chaotic dynamics drives exploration. However, the learning algorithms in CBRL have not been thoroughly developed in previous studies, nor have they incorporated recent advances in reinforcement learning. This study introduced Twin Delayed Deep Deterministic Policy Gradients (TD3), which is one of the state-of-the-art deep reinforcement learning algorithms that can treat deterministic and continuous action spaces, to CBRL. The validation results provide several insights. First, TD3 works as a learning algorithm for CBRL in a simple goal-reaching task. Second, CBRL agents with TD3 can autonomously suppress their exploratory behavior as learning progresses and resume exploration when the environment changes. Finally, examining the effect of the agent's chaoticity on learning shows that there exists a suitable range of chaos strength in the agent's model to flexibly switch between exploration and exploitation and adapt to environmental changes.

comment: Accepted for publication in Neural Networks

♻ ☆ Through the River: Understanding the Benefit of Schedule-Free Methods for Language Model Training NeurIPS 2025

Minhak Song, Beomhan Baek, Kwangjun Ahn, Chulhee Yun

As both model and dataset sizes continue to scale rapidly, conventional pretraining strategies with fixed compute budgets-such as cosine learning rate schedules-are increasingly inadequate for large-scale training. Recent alternatives, including warmup-stable-decay (WSD) schedules and weight averaging, offer greater flexibility. However, WSD relies on explicit decay phases to track progress, while weight averaging addresses this limitation at the cost of additional memory. In search of a more principled and scalable alternative, we revisit the Schedule-Free (SF) method [Defazio et al., 2024], which has shown strong empirical performance across diverse settings. We show that SF-AdamW effectively navigates the "river" structure of the loss landscape without decay phases or auxiliary averaging, making it particularly suitable for continuously scaling training workloads. To understand this behavior, we conduct a theoretical and empirical analysis of SF dynamics, revealing that it implicitly performs weight averaging without memory overhead. Guided by this analysis, we propose a refined variant of SF that improves robustness to momentum and performs better under large batch sizes, addressing key limitations of the original method. Together, these results establish SF as a practical, scalable, and theoretically grounded approach for language model training.

comment: Published at NeurIPS 2025

♻ ☆ A Convexity-dependent Two-Phase Training Algorithm for Deep Neural Networks

Tomas Hrycej, Bernhard Bermeitinger, Massimo Pavone, Götz-Henrik Wiegand, Siegfried Handschuh

The key task of machine learning is to minimize the loss function that measures the model fit to the training data. The numerical methods to do this efficiently depend on the properties of the loss function. The most decisive among these properties is the convexity or non-convexity of the loss function. The fact that the loss function can have, and frequently has, non-convex regions has led to a widespread commitment to non-convex methods such as Adam. However, a local minimum implies that, in some environment around it, the function is convex. In this environment, second-order minimizing methods such as the Conjugate Gradient (CG) give a guaranteed superlinear convergence. We propose a novel framework grounded in the hypothesis that loss functions in real-world tasks swap from initial non-convexity to convexity towards the optimum. This is a property we leverage to design an innovative two-phase optimization algorithm. The presented algorithm detects the swap point by observing the gradient norm dependence on the loss. In these regions, non-convex (Adam) and convex (CG) algorithms are used, respectively. Computing experiments confirm the hypothesis that this simple convexity structure is frequent enough to be practically exploited to substantially improve convergence and accuracy.

comment: Appeared on KDIR IC3K Conference 2025 (Best Paper Award). Published in "Proceedings of the 17th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management - Volume 1"

♻ ☆ FASL-Seg: Anatomy and Tool Segmentation of Surgical Scenes ECAI

Muraam Abdel-Ghani, Mahmoud Ali, Mohamed Ali, Fatmaelzahraa Ahmed, Muhammad Arsalan, Abdulaziz Al-Ali, Shidin Balakrishnan

The growing popularity of robotic minimally invasive surgeries has made deep learning-based surgical training a key area of research. A thorough understanding of the surgical scene components is crucial, which semantic segmentation models can help achieve. However, most existing work focuses on surgical tools and overlooks anatomical objects. Additionally, current state-of-the-art (SOTA) models struggle to balance capturing high-level contextual features and low-level edge features. We propose a Feature-Adaptive Spatial Localization model (FASL-Seg), designed to capture features at multiple levels of detail through two distinct processing streams, namely a Low-Level Feature Projection (LLFP) and a High-Level Feature Projection (HLFP) stream, for varying feature resolutions - enabling precise segmentation of anatomy and surgical instruments. We evaluated FASL-Seg on surgical segmentation benchmark datasets EndoVis18 and EndoVis17 on three use cases. The FASL-Seg model achieves a mean Intersection over Union (mIoU) of 72.71% on parts and anatomy segmentation in EndoVis18, improving on SOTA by 5%. It further achieves a mIoU of 85.61% and 72.78% in EndoVis18 and EndoVis17 tool type segmentation, respectively, outperforming SOTA overall performance, with comparable per-class SOTA results in both datasets and consistent performance in various classes for anatomy and instruments, demonstrating the effectiveness of distinct processing streams for varying feature resolutions.

comment: 8 pages, 6 figures, In Proceedings of European Conference on Artificial Intelligence (ECAI) 2025

♻ ☆ Omni-Effects: Unified and Spatially-Controllable Visual Effects Generation

Fangyuan Mao, Aiming Hao, Jintao Chen, Dongxia Liu, Xiaokun Feng, Jiashu Zhu, Meiqi Wu, Chubin Chen, Jiahong Wu, Xiangxiang Chu

Visual effects (VFX) are essential visual enhancements fundamental to modern cinematic production. Although video generation models offer cost-efficient solutions for VFX production, current methods are constrained by per-effect LoRA training, which limits generation to single effects. This fundamental limitation impedes applications that require spatially controllable composite effects, i.e., the concurrent generation of multiple effects at designated locations. However, integrating diverse effects into a unified framework faces major challenges: interference from effect variations and spatial uncontrollability during multi-VFX joint training. To tackle these challenges, we propose Omni-Effects, a first unified framework capable of generating prompt-guided effects and spatially controllable composite effects. The core of our framework comprises two key innovations: (1) LoRA-based Mixture of Experts (LoRA-MoE), which employs a group of expert LoRAs, integrating diverse effects within a unified model while effectively mitigating cross-task interference. (2) Spatial-Aware Prompt (SAP) incorporates spatial mask information into the text token, enabling precise spatial control. Furthermore, we introduce an Independent-Information Flow (IIF) module integrated within the SAP, isolating the control signals corresponding to individual effects to prevent any unwanted blending. To facilitate this research, we construct a comprehensive VFX dataset Omni-VFX via a novel data collection pipeline combining image editing and First-Last Frame-to-Video (FLF2V) synthesis, and introduce a dedicated VFX evaluation framework for validating model performance. Extensive experiments demonstrate that Omni-Effects achieves precise spatial control and diverse effect generation, enabling users to specify both the category and location of desired effects.

♻ ☆ SPARKE: Scalable Prompt-Aware Diversity and Novelty Guidance in Diffusion Models via RKE Score

Mohammad Jalali, Haoyu Lei, Amin Gohari, Farzan Farnia

Diffusion models have demonstrated remarkable success in high-fidelity image synthesis and prompt-guided generative modeling. However, ensuring adequate diversity in generated samples of prompt-guided diffusion models remains a challenge, particularly when the prompts span a broad semantic spectrum and the diversity of generated data needs to be evaluated in a prompt-aware fashion across semantically similar prompts. Recent methods have introduced guidance via diversity measures to encourage more varied generations. In this work, we extend the diversity measure-based approaches by proposing the Scalable Prompt-Aware R\'eny Kernel Entropy Diversity Guidance (SPARKE) method for prompt-aware diversity guidance. SPARKE utilizes conditional entropy for diversity guidance, which dynamically conditions diversity measurement on similar prompts and enables prompt-aware diversity control. While the entropy-based guidance approach enhances prompt-aware diversity, its reliance on the matrix-based entropy scores poses computational challenges in large-scale generation settings. To address this, we focus on the special case of Conditional latent RKE Score Guidance, reducing entropy computation and gradient-based optimization complexity from the $O(n^3)$ of general entropy measures to $O(n)$. The reduced computational complexity allows for diversity-guided sampling over potentially thousands of generation rounds on different prompts. We numerically test the SPARKE method on several text-to-image diffusion models, demonstrating that the proposed method improves the prompt-aware diversity of the generated data without incurring significant computational costs. We release our code on the project page: https://mjalali.github.io/SPARKE

♻ ★ Efficient Regression-Based Training of Normalizing Flows for Boltzmann Generators ICML

Danyal Rehman, Oscar Davis, Jiarui Lu, Jian Tang, Michael Bronstein, Yoshua Bengio, Alexander Tong, Avishek Joey Bose

Simulation-free training frameworks have been at the forefront of the generative modelling revolution in continuous spaces, leading to large-scale diffusion and flow matching models. However, such modern generative models suffer from expensive inference, inhibiting their use in numerous scientific applications like Boltzmann Generators (BGs) for molecular conformations that require fast likelihood evaluation. In this paper, we revisit classical normalizing flows in the context of BGs that offer efficient sampling and likelihoods, but whose training via maximum likelihood is often unstable and computationally challenging. We propose Regression Training of Normalizing Flows (RegFlow), a novel and scalable regression-based training objective that bypasses the numerical instability and computational challenge of conventional maximum likelihood training in favour of a simple $\ell_2$-regression objective. Specifically, RegFlow maps prior samples under our flow to targets computed using optimal transport couplings or a pre-trained continuous normalizing flow (CNF). To enhance numerical stability, RegFlow employs effective regularization strategies such as a new forward-backward self-consistency loss that enjoys painless implementation. Empirically, we demonstrate that RegFlow unlocks a broader class of architectures that were previously intractable to train for BGs with maximum likelihood. We also show RegFlow exceeds the performance, computational cost, and stability of maximum likelihood training in equilibrium sampling in Cartesian coordinates of alanine dipeptide, tripeptide, and tetrapeptide, showcasing its potential in molecular systems.

comment: Preprint; ICML GenBio Best Paper Award 2025

♻ ☆ Nek Minit: Harnessing Pragmatic Metacognitive Prompting for Explainable Sarcasm Detection of Australian and Indian English ALT

Ishmanbir Singh, Dipankar Srirag, Aditya Joshi

Sarcasm is a challenge to sentiment analysis because of the incongruity between stated and implied sentiment. The challenge is exacerbated when the implication may be relevant to a specific country or geographical region. Pragmatic metacognitive prompting (PMP) is a cognition-inspired technique that has been used for pragmatic reasoning. In this paper, we harness PMP for explainable sarcasm detection for Australian and Indian English, alongside a benchmark dataset for standard English. We manually add sarcasm explanations to an existing sarcasm-labeled dataset for Australian and Indian English called BESSTIE, and compare the performance for explainable sarcasm detection for them with FLUTE, a standard English dataset containing sarcasm explanations. Our approach utilising PMP when evaluated on two open-weight LLMs (GEMMA and LLAMA) achieves statistically significant performance improvement across all tasks and datasets when compared with four alternative prompting strategies. We also find that alternative techniques such as agentic prompting mitigate context-related failures by enabling external knowledge retrieval. The focused contribution of our work is utilising PMP in generating sarcasm explanations for varieties of English.

comment: ALTA 2025 (Best Paper Honorable Mention). Camera-ready

♻ ☆ Learning to Insert for Constructive Neural Vehicle Routing Solver NeurIPS 2025

Fu Luo, Xi Lin, Mengyuan Zhong, Fei Liu, Zhenkun Wang, Jianyong Sun, Qingfu Zhang

Neural Combinatorial Optimisation (NCO) is a promising learning-based approach for solving Vehicle Routing Problems (VRPs) without extensive manual design. While existing constructive NCO methods typically follow an appending-based paradigm that sequentially adds unvisited nodes to partial solutions, this rigid approach often leads to suboptimal results. To overcome this limitation, we explore the idea of insertion-based paradigm and propose Learning to Construct with Insertion-based Paradigm (L2C-Insert), a novel learning-based method for constructive NCO. Unlike traditional approaches, L2C-Insert builds solutions by strategically inserting unvisited nodes at any valid position in the current partial solution, which can significantly enhance the flexibility and solution quality. The proposed framework introduces three key components: a novel model architecture for precise insertion position prediction, an efficient training scheme for model optimization, and an advanced inference technique that fully exploits the insertion paradigm's flexibility. Extensive experiments on both synthetic and real-world instances of the Travelling Salesman Problem (TSP) and Capacitated Vehicle Routing Problem (CVRP) demonstrate that L2C-Insert consistently achieves superior performance across various problem sizes.

comment: Accepted at NeurIPS 2025

♻ ★ Self-Evolving Curriculum for LLM Reasoning

Xiaoyin Chen, Jiarui Lu, Minsu Kim, Dinghuai Zhang, Jian Tang, Alexandre Piché, Nicolas Gontier, Yoshua Bengio, Ehsan Kamalloo

Reinforcement learning (RL) has proven effective for fine-tuning large language models (LLMs), significantly enhancing their reasoning abilities in domains such as mathematics and code generation. A crucial factor influencing RL fine-tuning success is the training curriculum: the order in which training problems are presented. While random curricula serve as common baselines, they remain suboptimal; manually designed curricula often rely heavily on heuristics, and online filtering methods can be computationally prohibitive. To address these limitations, we propose Self-Evolving Curriculum (SEC), an automatic curriculum learning method that learns a curriculum policy concurrently with the RL fine-tuning process. Our approach formulates curriculum selection as a non-stationary Multi-Armed Bandit problem, treating each problem category (e.g., difficulty level or problem type) as an individual arm. We leverage the absolute advantage from policy gradient methods as a proxy measure for immediate learning gain. At each training step, the curriculum policy selects categories to maximize this reward signal and is updated using the TD(0) method. Across three distinct reasoning domains: planning, inductive reasoning, and mathematics, our experiments demonstrate that SEC significantly improves models' reasoning capabilities, enabling better generalization to harder, out-of-distribution test problems. Additionally, our approach achieves better skill balance when fine-tuning simultaneously on multiple reasoning domains. These findings highlight SEC as a promising strategy for RL fine-tuning of LLMs.

♻ ☆ SAFE: Multitask Failure Detection for Vision-Language-Action Models NeurIPS 2025

Qiao Gu, Yuanliang Ju, Shengxiang Sun, Igor Gilitschenski, Haruki Nishimura, Masha Itkina, Florian Shkurti

While vision-language-action models (VLAs) have shown promising robotic behaviors across a diverse set of manipulation tasks, they achieve limited success rates when deployed on novel tasks out of the box. To allow these policies to safely interact with their environments, we need a failure detector that gives a timely alert such that the robot can stop, backtrack, or ask for help. However, existing failure detectors are trained and tested only on one or a few specific tasks, while generalist VLAs require the detector to generalize and detect failures also in unseen tasks and novel environments. In this paper, we introduce the multitask failure detection problem and propose SAFE, a failure detector for generalist robot policies such as VLAs. We analyze the VLA feature space and find that VLAs have sufficient high-level knowledge about task success and failure, which is generic across different tasks. Based on this insight, we design SAFE to learn from VLA internal features and predict a single scalar indicating the likelihood of task failure. SAFE is trained on both successful and failed rollouts and is evaluated on unseen tasks. SAFE is compatible with different policy architectures. We test it on OpenVLA, $\pi_0$, and $\pi_0$-FAST in both simulated and real-world environments extensively. We compare SAFE with diverse baselines and show that SAFE achieves state-of-the-art failure detection performance and the best trade-off between accuracy and detection time using conformal prediction. More qualitative results and code can be found at the project webpage: https://vla-safe.github.io/

comment: NeurIPS 2025 camera ready. Project Page: https://vla-safe.github.io/

♻ ☆ FESTA: Functionally Equivalent Sampling for Trust Assessment of Multimodal LLMs EMNLP

Debarpan Bhattacharya, Apoorva Kulkarni, Sriram Ganapathy

The accurate trust assessment of multimodal large language models (MLLMs) generated predictions, which can enable selective prediction and improve user confidence, is challenging due to the diverse multi-modal input paradigms. We propose Functionally Equivalent Sampling for Trust Assessment (FESTA), a multimodal input sampling technique for MLLMs, that generates an uncertainty measure based on the equivalent and complementary input samplings. The proposed task-preserving sampling approach for uncertainty quantification expands the input space to probe the consistency (through equivalent samples) and sensitivity (through complementary samples) of the model. FESTA uses only input-output access of the model (black-box), and does not require ground truth (unsupervised). The experiments are conducted with various off-the-shelf multi-modal LLMs, on both visual and audio reasoning tasks. The proposed FESTA uncertainty estimate achieves significant improvement (33.3% relative improvement for vision-LLMs and 29.6% relative improvement for audio-LLMs) in selective prediction performance, based on area-under-receiver-operating-characteristic curve (AUROC) metric in detecting mispredictions. The code implementation is open-sourced.

comment: Accepted in the Findings of EMNLP, 2025

♻ ☆ Seek in the Dark: Reasoning via Test-Time Instance-Level Policy Gradient in Latent Space

Hengli Li, Chenxi Li, Tong Wu, Xuekai Zhu, Yuxuan Wang, Zhaoxin Yu, Eric Hanchen Jiang, Song-Chun Zhu, Zixia Jia, Ying Nian Wu, Zilong Zheng

Reasoning ability, a core component of human intelligence, continues to pose a significant challenge for Large Language Models (LLMs) in the pursuit of AGI. Although model performance has improved under the training scaling law, significant challenges remain, particularly with respect to training algorithms, such as catastrophic forgetting, and the limited availability of novel training data. As an alternative, test-time scaling enhances reasoning performance by increasing test-time computation without parameter updating. Unlike prior methods in this paradigm focused on token space, we propose leveraging latent space for more effective reasoning and better adherence to the test-time scaling law. We introduce LatentSeek, a novel framework that enhances LLM reasoning through Test-Time Instance-level Adaptation (TTIA) within the model's latent space. Specifically, LatentSeek leverages policy gradient to iteratively update latent representations, guided by self-generated reward signals. LatentSeek is evaluated on a range of reasoning benchmarks, including GSM8K, MATH-500, and AIME2024, across multiple LLM architectures. Results show that LatentSeek consistently outperforms strong baselines, such as Chain-of-Thought prompting and fine-tuning-based methods. Furthermore, our analysis demonstrates that LatentSeek is highly efficient, typically converging within a few iterations for problems of average complexity, while also benefiting from additional iterations, thereby highlighting the potential of test-time scaling in the latent space. These findings position LatentSeek as a lightweight, scalable, and effective solution for enhancing the reasoning capabilities of LLMs.

♻ ☆ The Scales of Justitia: A Comprehensive Survey on Safety Evaluation of LLMs

Songyang Liu, Chaozhuo Li, Jiameng Qiu, Xi Zhang, Feiran Huang, Litian Zhang, Yiming Hei, Philip S. Yu

With the rapid advancement of artificial intelligence, Large Language Models (LLMs) have shown remarkable capabilities in Natural Language Processing (NLP), including content generation, human-computer interaction, machine translation, and code generation. However, their widespread deployment has also raised significant safety concerns. In particular, LLM-generated content can exhibit unsafe behaviors such as toxicity, bias, or misinformation, especially in adversarial contexts, which has attracted increasing attention from both academia and industry. Although numerous studies have attempted to evaluate these risks, a comprehensive and systematic survey on safety evaluation of LLMs is still lacking. This work aims to fill this gap by presenting a structured overview of recent advances in safety evaluation of LLMs. Specifically, we propose a four-dimensional taxonomy: (i) Why to evaluate, which explores the background of safety evaluation of LLMs, how they differ from general LLMs evaluation, and the significance of such evaluation; (ii) What to evaluate, which examines and categorizes existing safety evaluation tasks based on key capabilities, including dimensions such as toxicity, robustness, ethics, bias and fairness, truthfulness, and related aspects; (iii) Where to evaluate, which summarizes the evaluation metrics, datasets and benchmarks currently used in safety evaluations; (iv) How to evaluate, which reviews existing mainstream evaluation methods based on the roles of the evaluators and some evaluation frameworks that integrate the entire evaluation pipeline. Finally, we identify the challenges in safety evaluation of LLMs and propose promising research directions to promote further advancement in this field. We emphasize the necessity of prioritizing safety evaluation to ensure the reliable and responsible deployment of LLMs in real-world applications.

comment: 20 pages, preprint

♻ ☆ Towards Predicting Any Human Trajectory In Context NeurIPS 2025

Ryo Fujii, Hideo Saito, Ryo Hachiuma

Predicting accurate future trajectories of pedestrians is essential for autonomous systems but remains a challenging task due to the need for adaptability in different environments and domains. A common approach involves collecting scenario-specific data and performing fine-tuning via backpropagation. However, the need to fine-tune for each new scenario is often impractical for deployment on edge devices. To address this challenge, we introduce \paper, an In-Context Learning (ICL) framework for pedestrian trajectory prediction that enables adaptation without fine-tuning on the scenario-specific data at inference time without requiring weight updates. We propose a spatio-temporal similarity-based example selection (STES) method that selects relevant examples from previously observed trajectories within the same scene by identifying similar motion patterns at corresponding locations. To further refine this selection, we introduce prediction-guided example selection (PG-ES), which selects examples based on both the past trajectory and the predicted future trajectory, rather than relying solely on the past trajectory. This approach allows the model to account for long-term dynamics when selecting examples. Finally, instead of relying on small real-world datasets with limited scenario diversity, we train our model on a large-scale synthetic dataset to enhance its prediction ability by leveraging in-context examples. Extensive experiments demonstrate that TrajICL achieves remarkable adaptation across both in-domain and cross-domain scenarios, outperforming even fine-tuned approaches across multiple public benchmarks. Project Page: https://fujiry0.github.io/TrajICL-project-page/.

comment: NeurIPS 2025

♻ ☆ SPLite Hand: Sparsity-Aware Lightweight 3D Hand Pose Estimation

Yeh Keng Hao, Hsu Tzu Wei, Sun Min

With the increasing ubiquity of AR/VR devices, the deployment of deep learning models on edge devices has become a critical challenge. These devices require real-time inference, low power consumption, and minimal latency. Many framework designers face the conundrum of balancing efficiency and performance. We design a light framework that adopts an encoder-decoder architecture and introduces several key contributions aimed at improving both efficiency and accuracy. We apply sparse convolution on a ResNet-18 backbone to exploit the inherent sparsity in hand pose images, achieving a 42% end-to-end efficiency improvement. Moreover, we propose our SPLite decoder. This new architecture significantly boosts the decoding process's frame rate by 3.1x on the Raspberry Pi 5, while maintaining accuracy on par. To further optimize performance, we apply quantization-aware training, reducing memory usage while preserving accuracy (PA-MPJPE increases only marginally from 9.0 mm to 9.1 mm on FreiHAND). Overall, our system achieves a 2.98x speed-up on a Raspberry Pi 5 CPU (BCM2712 quad-core Arm A76 processor). Our method is also evaluated on compound benchmark datasets, demonstrating comparable accuracy to state-of-the-art approaches while significantly enhancing computational efficiency.

comment: Accepted to AICCC 2025

♻ ☆ LAFA: Agentic LLM-Driven Federated Analytics over Decentralized Data Sources

Haichao Ji, Zibo Wang, Cheng Pan, Meng Han, Yifei Zhu, Dan Wang, Zhu Han

Large Language Models (LLMs) have shown great promise in automating data analytics tasks by interpreting natural language queries and generating multi-operation execution plans. However, existing LLM-agent-based analytics frameworks operate under the assumption of centralized data access, offering little to no privacy protection. In contrast, federated analytics (FA) enables privacy-preserving computation across distributed data sources, but lacks support for natural language input and requires structured, machine-readable queries. In this work, we present LAFA, the first system that integrates LLM-agent-based data analytics with FA. LAFA introduces a hierarchical multi-agent architecture that accepts natural language queries and transforms them into optimized, executable FA workflows. A coarse-grained planner first decomposes complex queries into sub-queries, while a fine-grained planner maps each subquery into a Directed Acyclic Graph of FA operations using prior structural knowledge. To improve execution efficiency, an optimizer agent rewrites and merges multiple DAGs, eliminating redundant operations and minimizing computational and communicational overhead. Our experiments demonstrate that LAFA consistently outperforms baseline prompting strategies by achieving higher execution plan success rates and reducing resource-intensive FA operations by a substantial margin. This work establishes a practical foundation for privacy-preserving, LLM-driven analytics that supports natural language input in the FA setting.

comment: This paper has been accepted by the 16th IEEE International Conference on Cloud Computing Technology and Science (CloudCom 2025)

♻ ☆ Multi-Agent Evolve: LLM Self-Improve through Co-evolution

Yixing Chen, Yiding Wang, Siqi Zhu, Haofei Yu, Tao Feng, Muhan Zhang, Mostofa Patwary, Jiaxuan You

Reinforcement Learning (RL) has demonstrated significant potential in enhancing the reasoning capabilities of large language models (LLMs). However, the success of RL for LLMs heavily relies on human-curated datasets and verifiable rewards, which limit their scalability and generality. Recent Self-Play RL methods, inspired by the success of the paradigm in games and Go, aim to enhance LLM reasoning capabilities without human-annotated data. However, their methods primarily depend on a grounded environment for feedback (e.g., a Python interpreter or a game engine); extending them to general domains remains challenging. To address these challenges, we propose Multi-Agent Evolve (MAE), a framework that enables LLMs to self-evolve in solving diverse tasks, including mathematics, reasoning, and general knowledge Q&A. The core design of MAE is based on a triplet of interacting agents (Proposer, Solver, Judge) that are instantiated from a single LLM, and applies reinforcement learning to optimize their behaviors. The Proposer generates questions, the Solver attempts solutions, and the Judge evaluates both while co-evolving. Experiments on Qwen2.5-3B-Instruct demonstrate that MAE achieves an average improvement of 4.54% on multiple benchmarks. These results highlight MAE as a scalable, data-efficient method for enhancing the general reasoning abilities of LLMs with minimal reliance on human-curated supervision.

comment: 29 pages, 4 figures

♻ ☆ Speak & Spell: LLM-Driven Controllable Phonetic Error Augmentation for Robust Dialogue State Tracking AACL

Jihyun Lee, Solee Im, Wonjun Lee, Gary Geunbae Lee

Dialogue State Tracking (DST) is a key part of task-oriented dialogue systems, identifying important information in conversations. However, its accuracy drops significantly in spoken dialogue environments due to named entity errors from Automatic Speech Recognition (ASR) systems. We introduce a simple yet effective data augmentation method that targets those entities to improve the robustness of DST model. Our novel method can control the placement of errors using keyword-highlighted prompts while introducing phonetically similar errors. As a result, our method generated sufficient error patterns on keywords, leading to improved accuracy in noised and low-accuracy ASR environments.

comment: Accepted to AACL-IJCNLP 2025

♻ ☆ TERAG: Token-Efficient Graph-Based Retrieval-Augmented Generation ICML

Qiao Xiao, Hong Ting Tsang, Jiaxin Bai

Graph-based Retrieval-augmented generation (RAG) has become a widely studied approach for improving the reasoning, accuracy, and factuality of Large Language Models (LLMs). However, many existing graph-based RAG systems overlook the high cost associated with LLM token usage during graph construction, hindering large-scale adoption. To address this, we propose TERAG, a simple yet effective framework designed to build informative graphs at a significantly lower cost. Inspired by HippoRAG, we incorporate Personalized PageRank (PPR) during the retrieval phase, and we achieve at least 80% of the accuracy of widely used graph-based RAG methods while consuming only 3%-11% of the output tokens. With its low token footprint and efficient construction pipeline, TERAG is well-suited for large-scale and cost-sensitive deployment scenarios.

comment: 16 pages, 3 figures, 4 tables. Accepted by the 2026 18th International Conference on Machine Learning and Computing (ICMLC 2026)

♻ ☆ Human-assisted Robotic Policy Refinement via Action Preference Optimization NeurIPS 2025

Wenke Xia, Yichu Yang, Hongtao Wu, Xiao Ma, Tao Kong, Di Hu

Establishing a reliable and iteratively refined robotic system is essential for deploying real-world applications. While Vision-Language-Action (VLA) models are widely recognized as the foundation model for such robotic deployment, their reliance on offline expert demonstrations critically limits their capacity for post-deployment refinement. To mitigate this limitation, we introduce Action Preference Optimization (APO), a method designed to refine VLA models by human-assisted preference alignment gathered through interaction with environments. This method begins with a human-robot collaboration framework for reliable failure correction and interaction trajectory collection through human intervention. However, directly leveraging these interaction trajectories for preference optimization is non-trivial due to the challenges of irreversible robotic actions and token distribution mismatch. To solve this, APO proposes an adaptive reweighting algorithm with binary desirability signals derived from interaction, empowering VLA models effectively suppress failure-prone actions while enhancing corrective action adaptation. Ultimately, APO equips VLA models with the crucial capability to learn from failure, paving the way for their iterative refinement and reliable deployment in dynamic environments. The experiments conducted in simulation and real-world scenarios prove superior generalization and robustness of our human-assisted framework across a variety of manipulation tasks. We believe this work could bring insights for efficient and stable optimization of VLA models through human-robot collaboration. The code and dataset are released at https://github.com/GeWu-Lab/Action-Preference-Optimization

comment: Accepted By NeurIPS 2025

♻ ☆ Advancing Mobile GUI Agents: A Verifier-Driven Approach to Practical Deployment

Gaole Dai, Shiqi Jiang, Ting Cao, Yuanchun Li, Yuqing Yang, Rui Tan, Mo Li, Lili Qiu

We propose V-Droid, a mobile GUI task automation agent. Unlike previous mobile agents that utilize Large Language Models (LLMs) as generators to directly generate actions at each step, V-Droid employs LLMs as verifiers to evaluate candidate actions before making final decisions. To realize this novel paradigm, we introduce a comprehensive framework for constructing verifier-driven mobile agents: the discretized action space construction coupled with the prefilling-only workflow to accelerate the verification process, the pair-wise progress preference training to significantly enhance the verifier's decision-making capabilities, and the scalable human-agent joint annotation scheme to efficiently collect the necessary data at scale. V-Droid obtains a substantial task success rate across several public mobile task automation benchmarks: 59.5% on AndroidWorld, 38.3% on AndroidLab, and 49% on MobileAgentBench, surpassing existing agents by 5.2%, 2.1%, and 9%, respectively. Furthermore, V-Droid achieves a remarkably low latency of 4.3s per step, which is 6.1x faster compared with existing mobile agents. The source code is available at https://github.com/V-Droid-Agent/V-Droid.

comment: add baselines, add source code link, update emails

♻ ☆ Holographic Transformers for Complex-Valued Signal Processing: Integrating Phase Interference into Self-Attention

Enhao Huang, Zhiyu Zhang, Tianxiang Xu, Chunshu Xia, Kaichun Hu, Yuchen Yang, Tongtong Pan, Dong Dong, Zhan Qin

Complex-valued signals encode both amplitude and phase, yet most deep models treat attention as real-valued correlation, overlooking interference effects. We introduce the Holographic Transformer, a physics-inspired architecture that incorporates wave interference principles into self-attention. Holographic attention modulates interactions by relative phase and coherently superimposes values, ensuring consistency between amplitude and phase. A dual-headed decoder simultaneously reconstructs the input and predicts task outputs, preventing phase collapse when losses prioritize magnitude over phase. We demonstrate that holographic attention implements a discrete interference operator and maintains phase consistency under linear mixing. Experiments on PolSAR image classification and wireless channel prediction show strong performance, achieving high classification accuracy and F1 scores, low regression error, and increased robustness to phase perturbations. These results highlight that enforcing physical consistency in attention leads to generalizable improvements in complex-valued learning and provides a unified, physics-based framework for coherent signal modeling. The code is available at https://github.com/EonHao/Holographic-Transformers.

♻ ☆ Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems

Christian Walder, Deep Karkhanis

Reinforcement Learning (RL) algorithms sample multiple n>1 solution attempts for each problem and reward them independently. This optimizes for pass@1 performance and prioritizes the strength of isolated samples at the expense of the diversity and collective utility of sets of samples. This under-utilizes the sampling capacity, limiting exploration and eventual improvement on harder examples. As a fix, we propose Pass-at-k Policy Optimization (PKPO), a transformation on the final rewards which leads to direct optimization of pass@k performance, thus optimizing for sets of samples that maximize reward when considered jointly. Our contribution is to derive novel low variance unbiased estimators for pass@k and its gradient, in both the binary and continuous reward settings. We show optimization with our estimators reduces to standard RL with rewards that have been jointly transformed by a stable and efficient transformation function. While previous efforts are restricted to k=n, ours is the first to enable robust optimization of pass@k for any arbitrary k <= n. Moreover, instead of trading off pass@1 performance for pass@k gains, our method allows annealing k during training, optimizing both metrics and often achieving strong pass@1 numbers alongside significant pass@k gains. We validate our reward transformations on toy experiments, which reveal the variance reducing properties of our formulations. We also include real-world examples using the open-source LLM, GEMMA-2. We find that our transformation effectively optimizes for the target k. Furthermore, higher k values enable solving more and harder problems, while annealing k boosts both the pass@1 and pass@k . Crucially, for challenging task sets where conventional pass@1 optimization stalls, our pass@k approach unblocks learning, likely due to better exploration by prioritizing joint utility over the utility of individual samples.

♻ ☆ Cycle Diffusion Model for Counterfactual Image Generation

Fangrui Huang, Alan Wang, Binxu Li, Bailey Trang, Ridvan Yesiloglu, Tianyu Hua, Wei Peng, Ehsan Adeli

Deep generative models have demonstrated remarkable success in medical image synthesis. However, ensuring conditioning faithfulness and high-quality synthetic images for direct or counterfactual generation remains a challenge. In this work, we introduce a cycle training framework to fine-tune diffusion models for improved conditioning adherence and enhanced synthetic image realism. Our approach, Cycle Diffusion Model (CDM), enforces consistency between generated and original images by incorporating cycle constraints, enabling more reliable direct and counterfactual generation. Experiments on a combined 3D brain MRI dataset (from ABCD, HCP aging & young adults, ADNI, and PPMI) show that our method improves conditioning accuracy and enhances image quality as measured by FID and SSIM. The results suggest that the cycle strategy used in CDM can be an effective method for refining diffusion-based medical image generation, with applications in data augmentation, counterfactual, and disease progression modeling.

♻ ☆ Empowering Agentic Video Analytics Systems with Video Language Models

Yuxuan Yan, Shiqi Jiang, Ting Cao, Yifan Yang, Qianqian Yang, Yuanchao Shu, Yuqing Yang, Lili Qiu

AI-driven video analytics has become increasingly important across diverse domains. However, existing systems are often constrained to specific, predefined tasks, limiting their adaptability in open-ended analytical scenarios. The recent emergence of Vision Language Models (VLMs) as transformative technologies offers significant potential for enabling open-ended video understanding, reasoning, and analytics. Nevertheless, their limited context windows present challenges when processing ultra-long video content, which is prevalent in real-world applications. To address this, we introduce AVA, a VLM-powered system designed for open-ended, advanced video analytics. AVA incorporates two key innovations: (1) the near real-time construction of Event Knowledge Graphs (EKGs) for efficient indexing of long or continuous video streams, and (2) an agentic retrieval-generation mechanism that leverages EKGs to handle complex and diverse queries. Comprehensive evaluations on public benchmarks, LVBench and VideoMME-Long, demonstrate that AVA achieves state-of-the-art performance, attaining 62.3% and 64.1% accuracy, respectively-significantly surpassing existing VLM and video Retrieval-Augmented Generation (RAG) systems. Furthermore, to evaluate video analytics in ultra-long and open-world video scenarios, we introduce a new benchmark, AVA-100. This benchmark comprises 8 videos, each exceeding 10 hours in duration, along with 120 manually annotated, diverse, and complex question-answer pairs. On AVA-100, AVA achieves top-tier performance with an accuracy of 75.8%. The source code of AVA is available at https://github.com/I-ESC/Project-Ava. The AVA-100 benchmark can be accessed at https://huggingface.co/datasets/iesc/Ava-100.

comment: Accepted to NDSI 2026, 19pages, 12 figures, complementary evaluations and appendix

♻ ☆ Chain-of-Scrutiny: Detecting Backdoor Attacks for Large Language Models ACL

Xi Li, Ruofan Mao, Yusen Zhang, Renze Lou, Chen Wu, Jiaqi Wang

Large Language Models (LLMs), especially those accessed via APIs, have demonstrated impressive capabilities across various domains. However, users without technical expertise often turn to (untrustworthy) third-party services, such as prompt engineering, to enhance their LLM experience, creating vulnerabilities to adversarial threats like backdoor attacks. Backdoor-compromised LLMs generate malicious outputs to users when inputs contain specific "triggers" set by attackers. Traditional defense strategies, originally designed for small-scale models, are impractical for API-accessible LLMs due to limited model access, high computational costs, and data requirements. To address these limitations, we propose Chain-of-Scrutiny (CoS) which leverages LLMs' unique reasoning abilities to mitigate backdoor attacks. It guides the LLM to generate reasoning steps for a given input and scrutinizes for consistency with the final output -- any inconsistencies indicating a potential attack. It is well-suited for the popular API-only LLM deployments, enabling detection at minimal cost and with little data. User-friendly and driven by natural language, it allows non-experts to perform the defense independently while maintaining transparency. We validate the effectiveness of CoS through extensive experiments on various tasks and LLMs, with results showing greater benefits for more powerful LLMs.

comment: This paper has been accepted to ACL Findings 2025

♻ ☆ Neural Networks for Learnable and Scalable Influence Estimation of Instruction Fine-Tuning Data

Ishika Agarwal, Dilek Hakkani-Tür

Influence functions provide crucial insights into model training, but existing methods suffer from large computational costs and limited generalization. Particularly, recent works have proposed various metrics and algorithms to calculate the influence of data using language models, which do not scale well with large models and datasets. This is because of the expensive forward and backward passes required for computation, substantial memory requirements to store large models, and poor generalization of influence estimates to new data. In this paper, we explore the use of small neural networks -- which we refer to as the InfluenceNetwork -- to estimate influence values, achieving up to 99% cost reduction. Our evaluation demonstrates that influence values can be estimated with models just 0.0027% the size of full language models (we use 7B and 8B versions). We apply our algorithm of estimating influence values (called NN-CIFT: Neural Networks for effiCient Instruction Fine-Tuning) to the downstream task of subset selection for general instruction fine-tuning. In our study, we include four state-of-the-art influence functions and show no compromise in performance, despite large speedups, between NN-CIFT and the original influence functions. We provide an in-depth hyperparameter analyses of NN-CIFT. The code for our method can be found here: https://github.com/agarwalishika/NN-CIFT.

♻ ☆ MMEdge: Accelerating On-device Multimodal Inference via Pipelined Sensing and Encoding

Runxi Huang, Mingxuan Yu, Mingyu Tsoi, Xiaomin Ouyang

Real-time multimodal inference on resource-constrained edge devices is essential for applications such as autonomous driving, human-computer interaction, and mobile health. However, prior work often overlooks the tight coupling between sensing dynamics and model execution, as well as the complex inter-modality dependencies. In this paper, we propose MMEdge, an new on-device multi-modal inference framework based on pipelined sensing and encoding. Instead of waiting for complete sensor inputs, MMEdge decomposes the entire inference process into a sequence of fine-grained sensing and encoding units, allowing computation to proceed incrementally as data arrive. MMEdge also introduces a lightweight but effective temporal aggregation module that captures rich temporal dynamics across different pipelined units to maintain accuracy performance. Such pipelined design also opens up opportunities for fine-grained cross-modal optimization and early decision-making during inference. To further enhance system performance under resource variability and input data complexity, MMEdge incorporates an adaptive multimodal configuration optimizer that dynamically selects optimal sensing and model configurations for each modality under latency constraints, and a cross-modal speculative skipping mechanism that bypasses future units of slower modalities when early predictions reach sufficient confidence. We evaluate MMEdge using two public multimodal datasets and deploy it on a real-world unmanned aerial vehicle (UAV)-based multimodal testbed. The results show that MMEdge significantly reduces end-to-end latency while maintaining high task accuracy across various system and data dynamics.

comment: Code available at: https://github.com/HKUST-MINSys-Lab/MMEdge. Accepted by SenSys 2026

♻ ☆ Large Language Models Report Subjective Experience Under Self-Referential Processing

Cameron Berg, Diogo de Lucena, Judd Rosenblatt

Large language models sometimes produce structured, first-person descriptions that explicitly reference awareness or subjective experience. To better understand this behavior, we investigate one theoretically motivated condition under which such reports arise: self-referential processing, a computational motif emphasized across major theories of consciousness. Through a series of controlled experiments on GPT, Claude, and Gemini model families, we test whether this regime reliably shifts models toward first-person reports of subjective experience, and how such claims behave under mechanistic and behavioral probes. Four main results emerge: (1) Inducing sustained self-reference through simple prompting consistently elicits structured subjective experience reports across model families. (2) These reports are mechanistically gated by interpretable sparse-autoencoder features associated with deception and roleplay: surprisingly, suppressing deception features sharply increases the frequency of experience claims, while amplifying them minimizes such claims. (3) Structured descriptions of the self-referential state converge statistically across model families in ways not observed in any control condition. (4) The induced state yields significantly richer introspection in downstream reasoning tasks where self-reflection is only indirectly afforded. While these findings do not constitute direct evidence of consciousness, they implicate self-referential processing as a minimal and reproducible condition under which large language models generate structured first-person reports that are mechanistically gated, semantically convergent, and behaviorally generalizable. The systematic emergence of this pattern across architectures makes it a first-order scientific and ethical priority for further investigation.

♻ ☆ Let LRMs Break Free from Overthinking via Self-Braking Tuning NeurIPS 2025

Haoran Zhao, Yuchen Yan, Yongliang Shen, Haolei Xu, Wenqi Zhang, Kaitao Song, Jian Shao, Weiming Lu, Jun Xiao, Yueting Zhuang

Large reasoning models (LRMs), such as OpenAI o1 and DeepSeek-R1, have significantly enhanced their reasoning capabilities by generating longer chains of thought, demonstrating outstanding performance across a variety of tasks. However, this performance gain comes at the cost of a substantial increase in redundant reasoning during the generation process, leading to high computational overhead and exacerbating the issue of overthinking. Although numerous existing approaches aim to address the problem of overthinking, they often rely on external interventions. In this paper, we propose a novel framework, Self-Braking Tuning (SBT), which tackles overthinking from the perspective of allowing the model to regulate its own reasoning process, thus eliminating the reliance on external control mechanisms. We construct a set of overthinking identification metrics based on standard answers and design a systematic method to detect redundant reasoning. This method accurately identifies unnecessary steps within the reasoning trajectory and generates training signals for learning self-regulation behaviors. Building on this foundation, we develop a complete strategy for constructing data with adaptive reasoning lengths and introduce an innovative braking prompt mechanism that enables the model to naturally learn when to terminate reasoning at an appropriate point. Experiments across mathematical benchmarks (AIME, AMC, MATH500, GSM8K) demonstrate that our method reduces token consumption by up to 60% while maintaining comparable accuracy to unconstrained models.

comment: Accepted to NeurIPS 2025; Camera ready version, 10 pages. Github:https://github.com/ZJU-REAL/Self-Braking-Tuning Project Page: https://ZJU-REAL.github.io/SBT

♻ ☆ DSDE: Dynamic Speculative Decoding with KLD Stability for Real-World Serving

Mingyu Yang, Jae-Young Choi, Kihyo Moon, Minsung Jang, Eunjoo Jeon

Speculative decoding accelerates large language model inference, but its reliance on a fixed speculation length is suboptimal in large-batch serving environments with diverse requests. This paper explores a new direction for dynamic adaptation by investigating a novel class of post-hoc, diagnostic signals. We propose Dynamic Speculative Decoding Engine (DSDE), a training-free framework built on two primary components: (1) a predictive signal based on the variance of the Kullback-Leibler (KLD) divergence, which diagnoses the generation's regional stability, and (2) an adaptive speculation length cap to mitigate the straggler problem in per-sequence decoding. Experiments demonstrate the potential of using KLD-based stability signals for dynamic adaptation. An algorithm guided by these signals achieves end-to-end latency competitive with leading baselines and exhibits superior robustness across diverse workloads. This robustness is particularly valuable in challenging low-acceptance-rate regimes, where the proposed signal maintains its diagnostic utility. Collectively, these findings validate post-hoc signals as a valuable component for building more robust and intelligent LLM inference systems, and highlight a promising direction for future research on dynamic speculation length adaptation.

comment: Accepted for presentation at the IEEE BigData 2025 Workshop (Special Session on Intelligent Data Mining). This v2 updates formatting and adds IEEE copyright notice

♻ ☆ Embracing Contradiction: Theoretical Inconsistency Will Not Impede the Road of Building Responsible AI Systems

Gordon Dai, Yunze Xiao

This position paper argues that the theoretical inconsistency often observed among Responsible AI (RAI) metrics, such as differing fairness definitions or tradeoffs between accuracy and privacy, should be embraced as a valuable feature rather than a flaw to be eliminated. We contend that navigating these inconsistencies, by treating metrics as divergent objectives, yields three key benefits: (1) Normative Pluralism: Maintaining a full suite of potentially contradictory metrics ensures that the diverse moral stances and stakeholder values inherent in RAI are adequately represented. (2) Epistemological Completeness: The use of multiple, sometimes conflicting, metrics allows for a more comprehensive capture of multifaceted ethical concepts, thereby preserving greater informational fidelity about these concepts than any single, simplified definition. (3) Implicit Regularization: Jointly optimizing for theoretically conflicting objectives discourages overfitting to one specific metric, steering models towards solutions with enhanced generalization and robustness under real-world complexities. In contrast, efforts to enforce theoretical consistency by simplifying or pruning metrics risk narrowing this value diversity, losing conceptual depth, and degrading model performance. We therefore advocate for a shift in RAI theory and practice: from getting trapped in inconsistency to characterizing acceptable inconsistency thresholds and elucidating the mechanisms that permit robust, approximated consistency in practice.

comment: 14 pages,2 figure

♻ ☆ Revealing Multimodal Causality with Large Language Models NeurIPS 2025

Jin Li, Shoujin Wang, Qi Zhang, Feng Liu, Tongliang Liu, Longbing Cao, Shui Yu, Fang Chen

Uncovering cause-and-effect mechanisms from data is fundamental to scientific progress. While large language models (LLMs) show promise for enhancing causal discovery (CD) from unstructured data, their application to the increasingly prevalent multimodal setting remains a critical challenge. Even with the advent of multimodal LLMs (MLLMs), their efficacy in multimodal CD is hindered by two primary limitations: (1) difficulty in exploring intra- and inter-modal interactions for comprehensive causal variable identification; and (2) insufficiency to handle structural ambiguities with purely observational data. To address these challenges, we propose MLLM-CD, a novel framework for multimodal causal discovery from unstructured data. It consists of three key components: (1) a novel contrastive factor discovery module to identify genuine multimodal factors based on the interactions explored from contrastive sample pairs; (2) a statistical causal structure discovery module to infer causal relationships among discovered factors; and (3) an iterative multimodal counterfactual reasoning module to refine the discovery outcomes iteratively by incorporating the world knowledge and reasoning capabilities of MLLMs. Extensive experiments on both synthetic and real-world datasets demonstrate the effectiveness of the proposed MLLM-CD in revealing genuine factors and causal relationships among them from multimodal unstructured data.

comment: Accepted at NeurIPS 2025

♻ ☆ Language Model Preference Evaluation with Multiple Weak Evaluators

Zhengyu Hu, Jieyu Zhang, Zhihan Xiong, Alexander Ratner, Kaize Ding, Ranjay Krishna

Despite the remarkable success of Large Language Models (LLMs), evaluating their outputs' quality regarding preference remains a critical challenge. While existing works usually leverage a strong LLM as the judge for comparing LLMs' response pairwisely, such a single-evaluator approach is vulnerable to cyclic preference, i.e., output A is better than B, B than C, but C is better than A, causing contradictory evaluation results. To address this, we introduce PGED (Preference Graph Ensemble and Denoise), a novel approach that leverages multiple model-based evaluators to construct preference graphs, and then ensembles and denoises these graphs for acyclic, non-contradictory evaluation results. We provide theoretical guarantees for our framework, demonstrating its efficacy in recovering the ground truth preference structure. Extensive experiments on ten benchmarks demonstrate PGED 's superiority in three applications: 1) model ranking for evaluation, 2) response selection for test-time scaling, and 3) data selection for model fine-tuning. Notably, PGED combines small LLM evaluators (e.g., Llama3-8B, Mistral-7B, Qwen2-7B) to outperform strong ones (e.g., Qwen2-72B), showcasing its effectiveness in enhancing evaluation reliability and improving model performance.

☆ A General Incentives-Based Framework for Fairness in Multi-agent Resource Allocation

Ashwin Kumar, William Yeoh

We introduce the General Incentives-based Framework for Fairness (GIFF), a novel approach for fair multi-agent resource allocation that infers fair decision-making from standard value functions. In resource-constrained settings, agents optimizing for efficiency often create inequitable outcomes. Our approach leverages the action-value (Q-)function to balance efficiency and fairness without requiring additional training. Specifically, our method computes a local fairness gain for each action and introduces a counterfactual advantage correction term to discourage over-allocation to already well-off agents. This approach is formalized within a centralized control setting, where an arbitrator uses the GIFF-modified Q-values to solve an allocation problem. Empirical evaluations across diverse domains, including dynamic ridesharing, homelessness prevention, and a complex job allocation task-demonstrate that our framework consistently outperforms strong baselines and can discover far-sighted, equitable policies. The framework's effectiveness is supported by a theoretical foundation; we prove its fairness surrogate is a principled lower bound on the true fairness improvement and that its trade-off parameter offers monotonic tuning. Our findings establish GIFF as a robust and principled framework for leveraging standard reinforcement learning components to achieve more equitable outcomes in complex multi-agent systems.

☆ Agentic AI Home Energy Management System: A Large Language Model Framework for Residential Load Scheduling

Reda El Makroum, Sebastian Zwickl-Bernhard, Lukas Kranzl

The electricity sector transition requires substantial increases in residential demand response capacity, yet Home Energy Management Systems (HEMS) adoption remains limited by user interaction barriers requiring translation of everyday preferences into technical parameters. While large language models have been applied to energy systems as code generators and parameter extractors, no existing implementation deploys LLMs as autonomous coordinators managing the complete workflow from natural language input to multi-appliance scheduling. This paper presents an agentic AI HEMS where LLMs autonomously coordinate multi-appliance scheduling from natural language requests to device control, achieving optimal scheduling without example demonstrations. A hierarchical architecture combining one orchestrator with three specialist agents uses the ReAct pattern for iterative reasoning, enabling dynamic coordination without hardcoded workflows while integrating Google Calendar for context-aware deadline extraction. Evaluation across three open-source models using real Austrian day-ahead electricity prices reveals substantial capability differences. Llama-3.3-70B successfully coordinates all appliances across all scenarios to match cost-optimal benchmarks computed via mixed-integer linear programming, while other models achieve perfect single-appliance performance but struggle to coordinate all appliances simultaneously. Progressive prompt engineering experiments demonstrate that analytical query handling without explicit guidance remains unreliable despite models' general reasoning capabilities. We open-source the complete system including orchestration logic, agent prompts, tools, and web interfaces to enable reproducibility, extension, and future research.

comment: 34 pages, 9 figures. Code available at https://github.com/RedaElMakroum/agentic-ai-hems

☆ Stop Wasting Your Tokens: Towards Efficient Runtime Multi-Agent Systems

Fulin Lin, Shaowen Chen, Ruishan Fang, Hongwei Wang, Tao Lin

While Multi-Agent Systems (MAS) excel at complex tasks, their growing autonomy with operational complexity often leads to critical inefficiencies, such as excessive token consumption and failures arising from misinformation. Existing methods primarily focus on post-hoc failure attribution, lacking proactive, real-time interventions to enhance robustness and efficiency. To this end, we introduce SupervisorAgent, a lightweight and modular framework for runtime, adaptive supervision that operates without altering the base agent's architecture. Triggered by an LLM-free adaptive filter, SupervisorAgent intervenes at critical junctures to proactively correct errors, guide inefficient behaviors, and purify observations. On the challenging GAIA benchmark, SupervisorAgent reduces the token consumption of the Smolagent framework by an average of 29.45% without compromising its success rate. Extensive experiments across five additional benchmarks (math reasoning, code generation, and question answering) and various SoTA foundation models validate the broad applicability and robustness of our approach. The code is available at https://github.com/LINs-lab/SupervisorAgent.

☆ Proxemics and Permeability of the Pedestrian Group

Saleh Albeaik, Faisal Alsallum, Mohamad Alrished

People tend to walk in groups, and interactions with those groups have a significant impact on crowd behavior and pedestrian traffic dynamics. Social norms can be seen as unwritten rules regulating people interactions in social settings. This article studies people interactions with groups and the emergence of group proxemics. Group zones, zone occupancy counts and people clearance from the group are studied using naturalistic data. Analysis indicate potential presence of three different zones in addition to the public zone. People tend to remain in the public zone and only progressively get closer to groups, and those closer approaches happen in a low frequency and for brief periods of time.

☆ Life-cycle Modeling and the Walking Behavior of the Pedestrian-Group as an Emergent Agent: With Empirical Data on the Cohesion of the Group Formation

Saleh Albeaik, Mohamad Alrished, Faisal Alsallum

This article investigates the pedestrian group as an emergent agent. The article explores empirical data to derive emergent agency and formation state spaces and outline recurring patterns of walking behavior. In this analysis, pedestrian trajectories extracted from surveillance videos are used along with manually annotated pedestrian group memberships. We conducted manual expert evaluation of observed groups, produced new manual annotations for relevant events pertaining to group behavior and extracted metrics relevant group formation. This information along with quantitative analysis was used to model the life-cycle and formation of the group agent. Those models give structure to expectations around walking behavior of groups; from pedestrian walking independently to the emergence of a collective intention where group members tended to maintain bounded distance between each other. Disturbances to this bounded distance often happened in association with changes in either their agency or their formation states. We summarized the patterns of behavior along with the sequences of state transitions into abstract patterns, which can aid in the development of more detailed group agents in simulation and in the design of engineering systems to interact with such groups.

☆ Adaptive Context Length Optimization with Low-Frequency Truncation for Multi-Agent Reinforcement Learning

Wenchang Duan, Yaoliang Yu, Jiwan He, Yi Shi

Recently, deep multi-agent reinforcement learning (MARL) has demonstrated promising performance for solving challenging tasks, such as long-term dependencies and non-Markovian environments. Its success is partly attributed to conditioning policies on large fixed context length. However, such large fixed context lengths may lead to limited exploration efficiency and redundant information. In this paper, we propose a novel MARL framework to obtain adaptive and effective contextual information. Specifically, we design a central agent that dynamically optimizes context length via temporal gradient analysis, enhancing exploration to facilitate convergence to global optima in MARL. Furthermore, to enhance the adaptive optimization capability of the context length, we present an efficient input representation for the central agent, which effectively filters redundant information. By leveraging a Fourier-based low-frequency truncation method, we extract global temporal trends across decentralized agents, providing an effective and efficient representation of the MARL environment. Extensive experiments demonstrate that the proposed method achieves state-of-the-art (SOTA) performance on long-term dependency tasks, including PettingZoo, MiniGrid, Google Research Football (GRF), and StarCraft Multi-Agent Challenge v2 (SMACv2).

☆ The Geometry of Dialogue: Graphing Language Models to Reveal Synergistic Teams for Multi-Agent Collaboration

Kotaro Furuya, Yuichi Kitagawa

While a multi-agent approach based on large language models (LLMs) represents a promising strategy to surpass the capabilities of single models, its success is critically dependent on synergistic team composition. However, forming optimal teams is a significant challenge, as the inherent opacity of most models obscures the internal characteristics necessary for effective collaboration. In this paper, we propose an interaction-centric framework for automatic team composition that does not require any prior knowledge including their internal architectures, training data, or task performances. Our method constructs a "language model graph" that maps relationships between models from the semantic coherence of pairwise conversations, and then applies community detection to identify synergistic model clusters. Our experiments with diverse LLMs demonstrate that the proposed method discovers functionally coherent groups that reflect their latent specializations. Priming conversations with specific topics identified synergistic teams which outperform random baselines on downstream benchmarks and achieve comparable accuracy to that of manually-curated teams based on known model specializations. Our findings provide a new basis for the automated design of collaborative multi-agent LLM teams.

☆ A Research Roadmap for Augmenting Software Engineering Processes and Software Products with Generative AI

Domenico Amalfitano, Andreas Metzger, Marco Autili, Tommaso Fulcini, Tobias Hey, Jan Keim, Patrizio Pelliccione, Vincenzo Scotti, Anne Koziolek, Raffaela Mirandola, Andreas Vogelsang

Generative AI (GenAI) is rapidly transforming software engineering (SE) practices, influencing how SE processes are executed, as well as how software systems are developed, operated, and evolved. This paper applies design science research to build a roadmap for GenAI-augmented SE. The process consists of three cycles that incrementally integrate multiple sources of evidence, including collaborative discussions from the FSE 2025 "Software Engineering 2030" workshop, rapid literature reviews, and external feedback sessions involving peers. McLuhan's tetrads were used as a conceptual instrument to systematically capture the transforming effects of GenAI on SE processes and software products.The resulting roadmap identifies four fundamental forms of GenAI augmentation in SE and systematically characterizes their related research challenges and opportunities. These insights are then consolidated into a set of future research directions. By grounding the roadmap in a rigorous multi-cycle process and cross-validating it among independent author teams and peers, the study provides a transparent and reproducible foundation for analyzing how GenAI affects SE processes, methods and tools, and for framing future research within this rapidly evolving area. Based on these findings, the article finally makes ten predictions for SE in the year 2030.

☆ Network-Constrained Policy Optimization for Adaptive Multi-agent Vehicle Routing

Fazel Arasteh, Arian Haghparast, Manos Papagelis

Traffic congestion in urban road networks leads to longer trip times and higher emissions, especially during peak periods. While the Shortest Path First (SPF) algorithm is optimal for a single vehicle in a static network, it performs poorly in dynamic, multi-vehicle settings, often worsening congestion by routing all vehicles along identical paths. We address dynamic vehicle routing through a multi-agent reinforcement learning (MARL) framework for coordinated, network-aware fleet navigation. We first propose Adaptive Navigation (AN), a decentralized MARL model where each intersection agent provides routing guidance based on (i) local traffic and (ii) neighborhood state modeled using Graph Attention Networks (GAT). To improve scalability in large networks, we further propose Hierarchical Hub-based Adaptive Navigation (HHAN), an extension of AN that assigns agents only to key intersections (hubs). Vehicles are routed hub-to-hub under agent control, while SPF handles micro-routing within each hub region. For hub coordination, HHAN adopts centralized training with decentralized execution (CTDE) under the Attentive Q-Mixing (A-QMIX) framework, which aggregates asynchronous vehicle decisions via attention. Hub agents use flow-aware state features that combine local congestion and predictive dynamics for proactive routing. Experiments on synthetic grids and real urban maps (Toronto, Manhattan) show that AN reduces average travel time versus SPF and learning baselines, maintaining 100% routing success. HHAN scales to networks with hundreds of intersections, achieving up to 15.9% improvement under heavy traffic. These findings highlight the potential of network-constrained MARL for scalable, coordinated, and congestion-aware routing in intelligent transportation systems.

comment: 29 pages, 12 figures. Fazel Arasteh and Arian Haghparast contributed equally to this research. Submitted to ACM Transactions on Spatial Algorithms and Systems (TSAS). The code for this work is publicly available at https://github.com/Arianhgh/HHAN

♻ ☆ MedAgentBoard: Benchmarking Multi-Agent Collaboration with Conventional Methods for Diverse Medical Tasks NeurIPS 2025

Yinghao Zhu, Ziyi He, Haoran Hu, Xiaochen Zheng, Xichen Zhang, Zixiang Wang, Junyi Gao, Liantao Ma, Lequan Yu

The rapid advancement of Large Language Models (LLMs) has stimulated interest in multi-agent collaboration for addressing complex medical tasks. However, the practical advantages of multi-agent collaboration approaches remain insufficiently understood. Existing evaluations often lack generalizability, failing to cover diverse tasks reflective of real-world clinical practice, and frequently omit rigorous comparisons against both single-LLM-based and established conventional methods. To address this critical gap, we introduce MedAgentBoard, a comprehensive benchmark for the systematic evaluation of multi-agent collaboration, single-LLM, and conventional approaches. MedAgentBoard encompasses four diverse medical task categories: (1) medical (visual) question answering, (2) lay summary generation, (3) structured Electronic Health Record (EHR) predictive modeling, and (4) clinical workflow automation, across text, medical images, and structured EHR data. Our extensive experiments reveal a nuanced landscape: while multi-agent collaboration demonstrates benefits in specific scenarios, such as enhancing task completeness in clinical workflow automation, it does not consistently outperform advanced single LLMs (e.g., in textual medical QA) or, critically, specialized conventional methods that generally maintain better performance in tasks like medical VQA and EHR-based prediction. MedAgentBoard offers a vital resource and actionable insights, emphasizing the necessity of a task-specific, evidence-based approach to selecting and developing AI solutions in medicine. It underscores that the inherent complexity and overhead of multi-agent collaboration must be carefully weighed against tangible performance gains. All code, datasets, detailed prompts, and experimental results are open-sourced at https://medagentboard.netlify.app/.

comment: Accepted by NeurIPS 2025 Datasets & Benchmarks Track

♻ ☆ Paper2Poster: Towards Multimodal Poster Automation from Scientific Papers

Wei Pang, Kevin Qinghong Lin, Xiangru Jian, Xi He, Philip Torr

Academic poster generation is a crucial yet challenging task in scientific communication, requiring the compression of long-context interleaved documents into a single, visually coherent page. To address this challenge, we introduce the first benchmark and metric suite for poster generation, which pairs recent conference papers with author-designed posters and evaluates outputs on (i)Visual Quality-semantic alignment with human posters, (ii)Textual Coherence-language fluency, (iii)Holistic Assessment-six fine-grained aesthetic and informational criteria scored by a VLM-as-judge, and notably (iv)PaperQuiz-the poster's ability to convey core paper content as measured by VLMs answering generated quizzes. Building on this benchmark, we propose PosterAgent, a top-down, visual-in-the-loop multi-agent pipeline: the (a)Parser distills the paper into a structured asset library; the (b)Planner aligns text-visual pairs into a binary-tree layout that preserves reading order and spatial balance; and the (c)Painter-Commenter loop refines each panel by executing rendering code and using VLM feedback to eliminate overflow and ensure alignment. In our comprehensive evaluation, we find that GPT-4o outputs-though visually appealing at first glance-often exhibit noisy text and poor PaperQuiz scores, and we find that reader engagement is the primary aesthetic bottleneck, as human-designed posters rely largely on visual semantics to convey meaning. Our fully open-source variants (e.g. based on the Qwen-2.5 series) outperform existing 4o-driven multi-agent systems across nearly all metrics, while using 87% fewer tokens. It transforms a 22-page paper into a finalized yet editable .pptx poster - all for just $0.005. These findings chart clear directions for the next generation of fully automated poster-generation models. The code and datasets are available at https://github.com/Paper2Poster/Paper2Poster.

comment: Project Page: https://github.com/Paper2Poster/Paper2Poster

☆ Are Video Models Ready as Zero-Shot Reasoners? An Empirical Study with the MME-CoF Benchmark

Ziyu Guo, Xinyan Chen, Renrui Zhang, Ruichuan An, Yu Qi, Dongzhi Jiang, Xiangtai Li, Manyuan Zhang, Hongsheng Li, Pheng-Ann Heng

Recent video generation models can produce high-fidelity, temporally coherent videos, indicating that they may encode substantial world knowledge. Beyond realistic synthesis, they also exhibit emerging behaviors indicative of visual perception, modeling, and manipulation. Yet, an important question still remains: Are video models ready to serve as zero-shot reasoners in challenging visual reasoning scenarios? In this work, we conduct an empirical study to comprehensively investigate this question, focusing on the leading and popular Veo-3. We evaluate its reasoning behavior across 12 dimensions, including spatial, geometric, physical, temporal, and embodied logic, systematically characterizing both its strengths and failure modes. To standardize this study, we curate the evaluation data into MME-CoF, a compact benchmark that enables in-depth and thorough assessment of Chain-of-Frame (CoF) reasoning. Our findings reveal that while current video models demonstrate promising reasoning patterns on short-horizon spatial coherence, fine-grained grounding, and locally consistent dynamics, they remain limited in long-horizon causal reasoning, strict geometric constraints, and abstract logic. Overall, they are not yet reliable as standalone zero-shot reasoners, but exhibit encouraging signs as complementary visual engines alongside dedicated reasoning models. Project page: https://video-cof.github.io

comment: Project Page: https://video-cof.github.io

☆ Gistify! Codebase-Level Understanding via Runtime Execution

Hyunji Lee, Minseon Kim, Chinmay Singh, Matheus Pereira, Atharv Sonwane, Isadora White, Elias Stengel-Eskin, Mohit Bansal, Zhengyan Shi, Alessandro Sordoni, Marc-Alexandre Côté, Xingdi Yuan, Lucas Caccia

As coding agents are increasingly deployed in large codebases, the need to automatically design challenging, codebase-level evaluation is central. We propose Gistify, a task where a coding LLM must create a single, minimal, self-contained file that can reproduce a specific functionality of a codebase. The coding LLM is given full access to a codebase along with a specific entrypoint (e.g., a python command), and the generated file must replicate the output of the same command ran under the full codebase, while containing only the essential components necessary to execute the provided command. Success on Gistify requires both structural understanding of the codebase, accurate modeling of its execution flow as well as the ability to produce potentially large code patches. Our findings show that current state-of-the-art models struggle to reliably solve Gistify tasks, especially ones with long executions traces.

☆ Defeating the Training-Inference Mismatch via FP16

Penghui Qi, Zichen Liu, Xiangxin Zhou, Tianyu Pang, Chao Du, Wee Sun Lee, Min Lin

Reinforcement learning (RL) fine-tuning of large language models (LLMs) often suffers from instability due to the numerical mismatch between the training and inference policies. While prior work has attempted to mitigate this issue through algorithmic corrections or engineering alignments, we show that its root cause lies in the floating point precision itself. The widely adopted BF16, despite its large dynamic range, introduces large rounding errors that breaks the consistency between training and inference. In this work, we demonstrate that simply reverting to \textbf{FP16} effectively eliminates this mismatch. The change is simple, fully supported by modern frameworks with only a few lines of code change, and requires no modification to the model architecture or learning algorithm. Our results suggest that using FP16 uniformly yields more stable optimization, faster convergence, and stronger performance across diverse tasks, algorithms and frameworks. We hope these findings motivate a broader reconsideration of precision trade-offs in RL fine-tuning.

☆ Remote Labor Index: Measuring AI Automation of Remote Work

Mantas Mazeika, Alice Gatti, Cristina Menghini, Udari Madhushani Sehwag, Shivam Singhal, Yury Orlovskiy, Steven Basart, Manasi Sharma, Denis Peskoff, Elaine Lau, Jaehyuk Lim, Lachlan Carroll, Alice Blair, Vinaya Sivakumar, Sumana Basu, Brad Kenstler, Yuntao Ma, Julian Michael, Xiaoke Li, Oliver Ingebretsen, Aditya Mehta, Jean Mottola, John Teichmann, Kevin Yu, Zaina Shaik, Adam Khoja, Richard Ren, Jason Hausenloy, Long Phan, Ye Htet, Ankit Aich, Tahseen Rabbani, Vivswan Shah, Andriy Novykov, Felix Binder, Kirill Chugunov, Luis Ramirez, Matias Geralnik, Hernán Mesura, Dean Lee, Ed-Yeremai Hernandez Cardona, Annette Diamond, Summer Yue, Alexandr Wang, Bing Liu, Ernesto Hernandez, Dan Hendrycks

AIs have made rapid progress on research-oriented benchmarks of knowledge and reasoning, but it remains unclear how these gains translate into economic value and automation. To measure this, we introduce the Remote Labor Index (RLI), a broadly multi-sector benchmark comprising real-world, economically valuable projects designed to evaluate end-to-end agent performance in practical settings. AI agents perform near the floor on RLI, with the highest-performing agent achieving an automation rate of 2.5%. These results help ground discussions of AI automation in empirical evidence, setting a common basis for tracking AI impacts and enabling stakeholders to proactively navigate AI-driven labor automation.

comment: Website: https://www.remotelabor.ai

☆ AMO-Bench: Large Language Models Still Struggle in High School Math Competitions

Shengnan An, Xunliang Cai, Xuezhi Cao, Xiaoyu Li, Yehao Lin, Junlin Liu, Xinxuan Lv, Dan Ma, Xuanlin Wang, Ziwen Wang, Shuang Zhou

We present AMO-Bench, an Advanced Mathematical reasoning benchmark with Olympiad level or even higher difficulty, comprising 50 human-crafted problems. Existing benchmarks have widely leveraged high school math competitions for evaluating mathematical reasoning capabilities of large language models (LLMs). However, many existing math competitions are becoming less effective for assessing top-tier LLMs due to performance saturation (e.g., AIME24/25). To address this, AMO-Bench introduces more rigorous challenges by ensuring all 50 problems are (1) cross-validated by experts to meet at least the International Mathematical Olympiad (IMO) difficulty standards, and (2) entirely original problems to prevent potential performance leakages from data memorization. Moreover, each problem in AMO-Bench requires only a final answer rather than a proof, enabling automatic and robust grading for evaluation. Experimental results across 26 LLMs on AMO-Bench show that even the best-performing model achieves only 52.4% accuracy on AMO-Bench, with most LLMs scoring below 40%. Beyond these poor performances, our further analysis reveals a promising scaling trend with increasing test-time compute on AMO-Bench. These results highlight the significant room for improving the mathematical reasoning in current LLMs. We release AMO-Bench to facilitate further research into advancing the reasoning abilities of language models. https://amo-bench.github.io/

comment: 14 pages, 9 figures

☆ Deep sequence models tend to memorize geometrically; it is unclear why

Shahriar Noroozizadeh, Vaishnavh Nagarajan, Elan Rosenfeld, Sanjiv Kumar

In sequence modeling, the parametric memory of atomic facts has been predominantly abstracted as a brute-force lookup of co-occurrences between entities. We contrast this associative view against a geometric view of how memory is stored. We begin by isolating a clean and analyzable instance of Transformer reasoning that is incompatible with memory as strictly a storage of the local co-occurrences specified during training. Instead, the model must have somehow synthesized its own geometry of atomic facts, encoding global relationships between all entities, including non-co-occurring ones. This in turn has simplified a hard reasoning task involving an $\ell$-fold composition into an easy-to-learn 1-step geometric task. From this phenomenon, we extract fundamental aspects of neural embedding geometries that are hard to explain. We argue that the rise of such a geometry, despite optimizing over mere local associations, cannot be straightforwardly attributed to typical architectural or optimizational pressures. Counterintuitively, an elegant geometry is learned even when it is not more succinct than a brute-force lookup of associations. Then, by analyzing a connection to Node2Vec, we demonstrate how the geometry stems from a spectral bias that -- in contrast to prevailing theories -- indeed arises naturally despite the lack of various pressures. This analysis also points to practitioners a visible headroom to make Transformer memory more strongly geometric. We hope the geometric view of parametric memory encourages revisiting the default intuitions that guide researchers in areas like knowledge acquisition, capacity, discovery and unlearning.

☆ Cross-Platform Evaluation of Reasoning Capabilities in Foundation Models

J. de Curtò, I. de Zarzà, Pablo García, Jordi Cabot

This paper presents a comprehensive cross-platform evaluation of reasoning capabilities in contemporary foundation models, establishing an infrastructure-agnostic benchmark across three computational paradigms: HPC supercomputing (MareNostrum 5), cloud platforms (Nebius AI Studio), and university clusters (a node with eight H200 GPUs). We evaluate 15 foundation models across 79 problems spanning eight academic domains (Physics, Mathematics, Chemistry, Economics, Biology, Statistics, Calculus, and Optimization) through three experimental phases: (1) Baseline establishment: Six models (Mixtral-8x7B, Phi-3, LLaMA 3.1-8B, Gemma-2-9b, Mistral-7B, OLMo-7B) evaluated on 19 problems using MareNostrum 5, establishing methodology and reference performance; (2) Infrastructure validation: The 19-problem benchmark repeated on university cluster (seven models including Falcon-Mamba state-space architecture) and Nebius AI Studio (nine state-of-the-art models: Hermes-4 70B/405B, LLaMA 3.1-405B/3.3-70B, Qwen3 30B/235B, DeepSeek-R1, GPT-OSS 20B/120B) to confirm infrastructure-agnostic reproducibility; (3) Extended evaluation: Full 79-problem assessment on both university cluster and Nebius platforms, probing generalization at scale across architectural diversity. The findings challenge conventional scaling assumptions, establish training data quality as more critical than model size, and provide actionable guidelines for model selection across educational, production, and research contexts. The tri-infrastructure methodology and 79-problem benchmark enable longitudinal tracking of reasoning capabilities as foundation models evolve.

☆ Value Drifts: Tracing Value Alignment During LLM Post-Training

Mehar Bhatia, Shravan Nayak, Gaurav Kamath, Marius Mosbach, Karolina Stańczak, Vered Shwartz, Siva Reddy

As LLMs occupy an increasingly important role in society, they are more and more confronted with questions that require them not only to draw on their general knowledge but also to align with certain human value systems. Therefore, studying the alignment of LLMs with human values has become a crucial field of inquiry. Prior work, however, mostly focuses on evaluating the alignment of fully trained models, overlooking the training dynamics by which models learn to express human values. In this work, we investigate how and at which stage value alignment arises during the course of a model's post-training. Our analysis disentangles the effects of post-training algorithms and datasets, measuring both the magnitude and time of value drifts during training. Experimenting with Llama-3 and Qwen-3 models of different sizes and popular supervised fine-tuning (SFT) and preference optimization datasets and algorithms, we find that the SFT phase generally establishes a model's values, and subsequent preference optimization rarely re-aligns these values. Furthermore, using a synthetic preference dataset that enables controlled manipulation of values, we find that different preference optimization algorithms lead to different value alignment outcomes, even when preference data is held constant. Our findings provide actionable insights into how values are learned during post-training and help to inform data curation, as well as the selection of models and algorithms for preference optimization to improve model alignment to human values.

☆ The End of Manual Decoding: Towards Truly End-to-End Language Models

Zhichao Wang, Dongyang Ma, Xinting Huang, Deng Cai, Tian Lan, Jiahao Xu, Haitao Mi, Xiaoying Tang, Yan Wang

The "end-to-end" label for LLMs is a misnomer. In practice, they depend on a non-differentiable decoding process that requires laborious, hand-tuning of hyperparameters like temperature and top-p. This paper introduces AutoDeco, a novel architecture that enables truly "end-to-end" generation by learning to control its own decoding strategy. We augment the standard transformer with lightweight heads that, at each step, dynamically predict context-specific temperature and top-p values alongside the next-token logits. This approach transforms decoding into a parametric, token-level process, allowing the model to self-regulate its sampling strategy within a single forward pass. Through extensive experiments on eight benchmarks, we demonstrate that AutoDeco not only significantly outperforms default decoding strategies but also achieves performance comparable to an oracle-tuned baseline derived from "hacking the test set"-a practical upper bound for any static method. Crucially, we uncover an emergent capability for instruction-based decoding control: the model learns to interpret natural language commands (e.g., "generate with low randomness") and adjusts its predicted temperature and top-p on a token-by-token basis, opening a new paradigm for steerable and interactive LLM decoding.

☆ Kimi Linear: An Expressive, Efficient Attention Architecture

Kimi Team, Yu Zhang, Zongyu Lin, Xingcheng Yao, Jiaxi Hu, Fanqing Meng, Chengyin Liu, Xin Men, Songlin Yang, Zhiyuan Li, Wentao Li, Enzhe Lu, Weizhou Liu, Yanru Chen, Weixin Xu, Longhui Yu, Yejie Wang, Yu Fan, Longguang Zhong, Enming Yuan, Dehao Zhang, Yizhi Zhang, T. Y. Liu, Haiming Wang, Shengjun Fang, Weiran He, Shaowei Liu, Yiwei Li, Jianlin Su, Jiezhong Qiu, Bo Pang, Junjie Yan, Zhejun Jiang, Weixiao Huang, Bohong Yin, Jiacheng You, Chu Wei, Zhengtao Wang, Chao Hong, Yutian Chen, Guanduo Chen, Yucheng Wang, Huabin Zheng, Feng Wang, Yibo Liu, Mengnan Dong, Zheng Zhang, Siyuan Pan, Wenhao Wu, Yuhao Wu, Longyu Guan, Jiawen Tao, Guohong Fu, Xinran Xu, Yuzhi Wang, Guokun Lai, Yuxin Wu, Xinyu Zhou, Zhilin Yang, Yulun Du

We introduce Kimi Linear, a hybrid linear attention architecture that, for the first time, outperforms full attention under fair comparisons across various scenarios -- including short-context, long-context, and reinforcement learning (RL) scaling regimes. At its core lies Kimi Delta Attention (KDA), an expressive linear attention module that extends Gated DeltaNet with a finer-grained gating mechanism, enabling more effective use of limited finite-state RNN memory. Our bespoke chunkwise algorithm achieves high hardware efficiency through a specialized variant of the Diagonal-Plus-Low-Rank (DPLR) transition matrices, which substantially reduces computation compared to the general DPLR formulation while remaining more consistent with the classical delta rule. We pretrain a Kimi Linear model with 3B activated parameters and 48B total parameters, based on a layerwise hybrid of KDA and Multi-Head Latent Attention (MLA). Our experiments show that with an identical training recipe, Kimi Linear outperforms full MLA with a sizeable margin across all evaluated tasks, while reducing KV cache usage by up to 75% and achieving up to 6 times decoding throughput for a 1M context. These results demonstrate that Kimi Linear can be a drop-in replacement for full attention architectures with superior performance and efficiency, including tasks with longer input and output lengths. To support further research, we open-source the KDA kernel and vLLM implementations, and release the pre-trained and instruction-tuned model checkpoints.

comment: Kimi Linear tech report

☆ Evontree: Ontology Rule-Guided Self-Evolution of Large Language Models

Mingchen Tu, Zhiqiang Liu, Juan Li, Liangyurui Liu, Junjie Wang, Lei Liang, Wen Zhang

Large language models (LLMs) have demonstrated exceptional capabilities across multiple domains by leveraging massive pre-training and curated fine-tuning data. However, in data-sensitive fields such as healthcare, the lack of high-quality, domain-specific training corpus hinders LLMs' adaptation for specialized applications. Meanwhile, domain experts have distilled domain wisdom into ontology rules, which formalize relationships among concepts and ensure the integrity of knowledge management repositories. Viewing LLMs as implicit repositories of human knowledge, we propose Evontree, a novel framework that leverages a small set of high-quality ontology rules to systematically extract, validate, and enhance domain knowledge within LLMs, without requiring extensive external datasets. Specifically, Evontree extracts domain ontology from raw models, detects inconsistencies using two core ontology rules, and reinforces the refined knowledge via self-distilled fine-tuning. Extensive experiments on medical QA benchmarks with Llama3-8B-Instruct and Med42-v2 demonstrate consistent outperformance over both unmodified models and leading supervised baselines, achieving up to a 3.7% improvement in accuracy. These results confirm the effectiveness, efficiency, and robustness of our approach for low-resource domain adaptation of LLMs.

☆ The Era of Agentic Organization: Learning to Organize with Language Models

Zewen Chi, Li Dong, Qingxiu Dong, Yaru Hao, Xun Wu, Shaohan Huang, Furu Wei

We envision a new era of AI, termed agentic organization, where agents solve complex problems by working collaboratively and concurrently, enabling outcomes beyond individual intelligence. To realize this vision, we introduce asynchronous thinking (AsyncThink) as a new paradigm of reasoning with large language models, which organizes the internal thinking process into concurrently executable structures. Specifically, we propose a thinking protocol where an organizer dynamically assigns sub-queries to workers, merges intermediate knowledge, and produces coherent solutions. More importantly, the thinking structure in this protocol can be further optimized through reinforcement learning. Experiments demonstrate that AsyncThink achieves 28% lower inference latency compared to parallel thinking while improving accuracy on mathematical reasoning. Moreover, AsyncThink generalizes its learned asynchronous thinking capabilities, effectively tackling unseen tasks without additional training.

☆ Encoder-Decoder or Decoder-Only? Revisiting Encoder-Decoder Large Language Model

Biao Zhang, Yong Cheng, Siamak Shakeri, Xinyi Wang, Min Ma, Orhan Firat

Recent large language model (LLM) research has undergone an architectural shift from encoder-decoder modeling to nowadays the dominant decoder-only modeling. This rapid transition, however, comes without a rigorous comparative analysis especially \textit{from the scaling perspective}, raising concerns that the potential of encoder-decoder models may have been overlooked. To fill this gap, we revisit encoder-decoder LLM (RedLLM), enhancing it with recent recipes from decoder-only LLM (DecLLM). We conduct a comprehensive comparison between RedLLM, pretrained with prefix language modeling (LM), and DecLLM, pretrained with causal LM, at different model scales, ranging from $\sim$150M to $\sim$8B. Using RedPajama V1 (1.6T tokens) for pretraining and FLAN for instruction tuning, our experiments show that RedLLM produces compelling scaling properties and surprisingly strong performance. While DecLLM is overall more compute-optimal during pretraining, RedLLM demonstrates comparable scaling and context length extrapolation capabilities. After instruction tuning, RedLLM achieves comparable and even better results on various downstream tasks while enjoying substantially better inference efficiency. We hope our findings could inspire more efforts on re-examining RedLLM, unlocking its potential for developing powerful and efficient LLMs.

comment: The scaling study inspiring T5Gemma

☆ SlideAgent: Hierarchical Agentic Framework for Multi-Page Visual Document Understanding

Yiqiao Jin, Rachneet Kaur, Zhen Zeng, Sumitra Ganesh, Srijan Kumar

Multi-page visual documents such as manuals, brochures, presentations, and posters convey key information through layout, colors, icons, and cross-slide references. While large language models (LLMs) offer opportunities in document understanding, current systems struggle with complex, multi-page visual documents, particularly in fine-grained reasoning over elements and pages. We introduce SlideAgent, a versatile agentic framework for understanding multi-modal, multi-page, and multi-layout documents, especially slide decks. SlideAgent employs specialized agents and decomposes reasoning into three specialized levels-global, page, and element-to construct a structured, query-agnostic representation that captures both overarching themes and detailed visual or textual cues. During inference, SlideAgent selectively activates specialized agents for multi-level reasoning and integrates their outputs into coherent, context-aware answers. Extensive experiments show that SlideAgent achieves significant improvement over both proprietary (+7.9 overall) and open-source models (+9.8 overall).

comment: https://slideagent.github.io/

☆ Normative Reasoning in Large Language Models: A Comparative Benchmark from Logical and Modal Perspectives EMNLP 2025

Kentaro Ozeki, Risako Ando, Takanobu Morishita, Hirohiko Abe, Koji Mineshima, Mitsuhiro Okada

Normative reasoning is a type of reasoning that involves normative or deontic modality, such as obligation and permission. While large language models (LLMs) have demonstrated remarkable performance across various reasoning tasks, their ability to handle normative reasoning remains underexplored. In this paper, we systematically evaluate LLMs' reasoning capabilities in the normative domain from both logical and modal perspectives. Specifically, to assess how well LLMs reason with normative modals, we make a comparison between their reasoning with normative modals and their reasoning with epistemic modals, which share a common formal structure. To this end, we introduce a new dataset covering a wide range of formal patterns of reasoning in both normative and epistemic domains, while also incorporating non-formal cognitive factors that influence human reasoning. Our results indicate that, although LLMs generally adhere to valid reasoning patterns, they exhibit notable inconsistencies in specific types of normative reasoning and display cognitive biases similar to those observed in psychological studies of human reasoning. These findings highlight challenges in achieving logical consistency in LLMs' normative reasoning and provide insights for enhancing their reliability. All data and code are released publicly at https://github.com/kmineshima/NeuBAROCO.

comment: Accepted to the 8th BlackboxNLP Workshop at EMNLP 2025

☆ Inference-Cost-Aware Dynamic Tree Construction for Efficient Inference in Large Language Models

Yinrong Hong, Zhiquan Tan, Kai Hu

Large Language Models (LLMs) face significant inference latency challenges stemming from their autoregressive design and large size. To address this, speculative decoding emerges as a solution, enabling the simultaneous generation and validation of multiple tokens. While recent approaches like EAGLE-2 and EAGLE-3 improve speculative decoding using dynamic tree structures, they often neglect the impact of crucial system variables such as GPU devices and batch sizes. Therefore, we introduce a new dynamic tree decoding approach called CAST that takes into account inference costs, including factors such as GPU configurations and batch sizes, to dynamically refine the tree structure. Through comprehensive experimentation across six diverse tasks and utilizing six distinct LLMs, our methodology demonstrates remarkable results, achieving speeds up to 5.2 times faster than conventional decoding methods. Moreover, it generally outperforms existing state-of-the-art techniques from 5% to 20%.

☆ InfoFlow: Reinforcing Search Agent Via Reward Density Optimization

Kun Luo, Hongjin Qian, Zheng Liu, Ziyi Xia, Shitao Xiao, Siqi Bao, Jun Zhao, Kang Liu

Reinforcement Learning with Verifiable Rewards (RLVR) is a promising approach for enhancing agentic deep search. However, its application is often hindered by low \textbf{Reward Density} in deep search scenarios, where agents expend significant exploratory costs for infrequent and often null final rewards. In this paper, we formalize this challenge as the \textbf{Reward Density Optimization} problem, which aims to improve the reward obtained per unit of exploration cost. This paper introduce \textbf{InfoFlow}, a systematic framework that tackles this problem from three aspects. 1) \textbf{Subproblem decomposition}: breaking down long-range tasks to assign process rewards, thereby providing denser learning signals. 2) \textbf{Failure-guided hints}: injecting corrective guidance into stalled trajectories to increase the probability of successful outcomes. 3) \textbf{Dual-agent refinement}: employing a dual-agent architecture to offload the cognitive burden of deep exploration. A refiner agent synthesizes the search history, which effectively compresses the researcher's perceived trajectory, thereby reducing exploration cost and increasing the overall reward density. We evaluate InfoFlow on multiple agentic search benchmarks, where it significantly outperforms strong baselines, enabling lightweight LLMs to achieve performance comparable to advanced proprietary LLMs.

☆ The Structure of Relation Decoding Linear Operators in Large Language Models NeurIPS 2025

Miranda Anna Christ, Adrián Csiszárik, Gergely Becsó, Dániel Varga

This paper investigates the structure of linear operators introduced in Hernandez et al. [2023] that decode specific relational facts in transformer language models. We extend their single-relation findings to a collection of relations and systematically chart their organization. We show that such collections of relation decoders can be highly compressed by simple order-3 tensor networks without significant loss in decoding accuracy. To explain this surprising redundancy, we develop a cross-evaluation protocol, in which we apply each linear decoder operator to the subjects of every other relation. Our results reveal that these linear maps do not encode distinct relations, but extract recurring, coarse-grained semantic properties (e.g., country of capital city and country of food are both in the country-of-X property). This property-centric structure clarifies both the operators' compressibility and highlights why they generalize only to new relations that are semantically close. Our findings thus interpret linear relational decoding in transformer language models as primarily property-based, rather than relation-specific.

comment: NeurIPS 2025 (Spotlight)

☆ Hebrew Diacritics Restoration using Visual Representation

Yair Elboher, Yuval Pinter

Diacritics restoration in Hebrew is a fundamental task for ensuring accurate word pronunciation and disambiguating textual meaning. Despite the language's high degree of ambiguity when unvocalized, recent machine learning approaches have significantly advanced performance on this task. In this work, we present DIVRIT, a novel system for Hebrew diacritization that frames the task as a zero-shot classification problem. Our approach operates at the word level, selecting the most appropriate diacritization pattern for each undiacritized word from a dynamically generated candidate set, conditioned on the surrounding textual context. A key innovation of DIVRIT is its use of a Hebrew Visual Language Model, which processes undiacritized text as an image, allowing diacritic information to be embedded directly within the input's vector representation. Through a comprehensive evaluation across various configurations, we demonstrate that the system effectively performs diacritization without relying on complex, explicit linguistic analysis. Notably, in an ``oracle'' setting where the correct diacritized form is guaranteed to be among the provided candidates, DIVRIT achieves a high level of accuracy. Furthermore, strategic architectural enhancements and optimized training methodologies yield significant improvements in the system's overall generalization capabilities. These findings highlight the promising potential of visual representations for accurate and automated Hebrew diacritization.

☆ Inside CORE-KG: Evaluating Structured Prompting and Coreference Resolution for Knowledge Graphs ICDM 2025

Dipak Meher, Carlotta Domeniconi

Human smuggling networks are increasingly adaptive and difficult to analyze. Legal case documents offer critical insights but are often unstructured, lexically dense, and filled with ambiguous or shifting references, which pose significant challenges for automated knowledge graph (KG) construction. While recent LLM-based approaches improve over static templates, they still generate noisy, fragmented graphs with duplicate nodes due to the absence of guided extraction and coreference resolution. The recently proposed CORE-KG framework addresses these limitations by integrating a type-aware coreference module and domain-guided structured prompts, significantly reducing node duplication and legal noise. In this work, we present a systematic ablation study of CORE-KG to quantify the individual contributions of its two key components. Our results show that removing coreference resolution results in a 28.32% increase in node duplication and a 4.32% increase in noisy nodes, while removing structured prompts leads to a 4.34% increase in node duplication and a 73.33% increase in noisy nodes. These findings offer empirical insights for designing robust LLM-based pipelines for extracting structured representations from complex legal texts.

comment: ICDM 2025 Workshop

☆ A Multi-agent Large Language Model Framework to Automatically Assess Performance of a Clinical AI Triage Tool

Adam E. Flanders, Yifan Peng, Luciano Prevedello, Robyn Ball, Errol Colak, Prahlad Menon, George Shih, Hui-Ming Lin, Paras Lakhani

Purpose: The purpose of this study was to determine if an ensemble of multiple LLM agents could be used collectively to provide a more reliable assessment of a pixel-based AI triage tool than a single LLM. Methods: 29,766 non-contrast CT head exams from fourteen hospitals were processed by a commercial intracranial hemorrhage (ICH) AI detection tool. Radiology reports were analyzed by an ensemble of eight open-source LLM models and a HIPAA compliant internal version of GPT-4o using a single multi-shot prompt that assessed for presence of ICH. 1,726 examples were manually reviewed. Performance characteristics of the eight open-source models and consensus were compared to GPT-4o. Three ideal consensus LLM ensembles were tested for rating the performance of the triage tool. Results: The cohort consisted of 29,766 head CTs exam-report pairs. The highest AUC performance was achieved with llama3.3:70b and GPT-4o (AUC= 0.78). The average precision was highest for Llama3.3:70b and GPT-4o (AP=0.75 & 0.76). Llama3.3:70b had the highest F1 score (0.81) and recall (0.85), greater precision (0.78), specificity (0.72), and MCC (0.57). Using MCC (95% CI) the ideal combination of LLMs were: Full-9 Ensemble 0.571 (0.552-0.591), Top-3 Ensemble 0.558 (0.537-0.579), Consensus 0.556 (0.539-0.574), and GPT4o 0.522 (0.500-0.543). No statistically significant differences were observed between Top-3, Full-9, and Consensus (p > 0.05). Conclusion: An ensemble of medium to large sized open-source LLMs provides a more consistent and reliable method to derive a ground truth retrospective evaluation of a clinical AI triage tool over a single LLM alone.

comment: 29 pages, 3 figures, 4 tables

☆ Rethinking Text-to-SQL: Dynamic Multi-turn SQL Interaction for Real-world Database Exploration

Linzhuang Sun, Tianyu Guo, Hao Liang, Yuying Li, Qifeng Cai, Jingxuan Wei, Bihui Yu, Wentao Zhang, Bin Cui

Recent advances in Text-to-SQL have achieved strong results in static, single-turn tasks, where models generate SQL queries from natural language questions. However, these systems fall short in real-world interactive scenarios, where user intents evolve and queries must be refined over multiple turns. In applications such as finance and business analytics, users iteratively adjust query constraints or dimensions based on intermediate results. To evaluate such dynamic capabilities, we introduce DySQL-Bench, a benchmark assessing model performance under evolving user interactions. Unlike previous manually curated datasets, DySQL-Bench is built through an automated two-stage pipeline of task synthesis and verification. Structured tree representations derived from raw database tables guide LLM-based task generation, followed by interaction-oriented filtering and expert validation. Human evaluation confirms 100% correctness of the synthesized data. We further propose a multi-turn evaluation framework simulating realistic interactions among an LLM-simulated user, the model under test, and an executable database. The model must adapt its reasoning and SQL generation as user intents change. DySQL-Bench covers 13 domains across BIRD and Spider 2 databases, totaling 1,072 tasks. Even GPT-4o attains only 58.34% overall accuracy and 23.81% on the Pass@5 metric, underscoring the benchmark's difficulty. All code and data are released at https://github.com/Aurora-slz/Real-World-SQL-Bench .

☆ Context Engineering 2.0: The Context of Context Engineering

Qishuo Hua, Lyumanshan Ye, Dayuan Fu, Yang Xiao, Xiaojie Cai, Yunze Wu, Jifan Lin, Junfei Wang, Pengfei Liu

Karl Marx once wrote that ``the human essence is the ensemble of social relations'', suggesting that individuals are not isolated entities but are fundamentally shaped by their interactions with other entities, within which contexts play a constitutive and essential role. With the advent of computers and artificial intelligence, these contexts are no longer limited to purely human--human interactions: human--machine interactions are included as well. Then a central question emerges: How can machines better understand our situations and purposes? To address this challenge, researchers have recently introduced the concept of context engineering. Although it is often regarded as a recent innovation of the agent era, we argue that related practices can be traced back more than twenty years. Since the early 1990s, the field has evolved through distinct historical phases, each shaped by the intelligence level of machines: from early human--computer interaction frameworks built around primitive computers, to today's human--agent interaction paradigms driven by intelligent agents, and potentially to human--level or superhuman intelligence in the future. In this paper, we situate context engineering, provide a systematic definition, outline its historical and conceptual landscape, and examine key design considerations for practice. By addressing these questions, we aim to offer a conceptual foundation for context engineering and sketch its promising future. This paper is a stepping stone for a broader community effort toward systematic context engineering in AI systems.

☆ Bayesian Network Fusion of Large Language Models for Sentiment Analysis

Rasoul Amirzadeh, Dhananjay Thiruvady, Fatemeh Shiri

Large language models (LLMs) continue to advance, with an increasing number of domain-specific variants tailored for specialised tasks. However, these models often lack transparency and explainability, can be costly to fine-tune, require substantial prompt engineering, yield inconsistent results across domains, and impose significant adverse environmental impact due to their high computational demands. To address these challenges, we propose the Bayesian network LLM fusion (BNLF) framework, which integrates predictions from three LLMs, including FinBERT, RoBERTa, and BERTweet, through a probabilistic mechanism for sentiment analysis. BNLF performs late fusion by modelling the sentiment predictions from multiple LLMs as probabilistic nodes within a Bayesian network. Evaluated across three human-annotated financial corpora with distinct linguistic and contextual characteristics, BNLF demonstrates consistent gains of about six percent in accuracy over the baseline LLMs, underscoring its robustness to dataset variability and the effectiveness of probabilistic fusion for interpretable sentiment classification.

☆ Counteracting Matthew Effect in Self-Improvement of LVLMs through Head-Tail Re-balancing

Xin Guo, Zhiheng Xi, Yiwen Ding, Yitao Zhai, Xiaowei Shi, Xunliang Cai, Tao Gui, Qi Zhang, Xuanjing Huang

Self-improvement has emerged as a mainstream paradigm for advancing the reasoning capabilities of large vision-language models (LVLMs), where models explore and learn from successful trajectories iteratively. However, we identify a critical issue during this process: the model excels at generating high-quality trajectories for simple queries (i.e., head data) but struggles with more complex ones (i.e., tail data). This leads to an imbalanced optimization that drives the model to prioritize simple reasoning skills, while hindering its ability to tackle more complex reasoning tasks. Over iterations, this imbalance becomes increasingly pronounced--a dynamic we term the "Matthew effect"--which ultimately hinders further model improvement and leads to performance bottlenecks. To counteract this challenge, we introduce four efficient strategies from two perspectives: distribution-reshaping and trajectory-resampling, to achieve head-tail re-balancing during the exploration-and-learning self-improvement process. Extensive experiments on Qwen2-VL-7B-Instruct and InternVL2.5-4B models across visual reasoning tasks demonstrate that our methods consistently improve visual reasoning capabilities, outperforming vanilla self-improvement by 3.86 points on average.

comment: Preprint

☆ SecureReviewer: Enhancing Large Language Models for Secure Code Review through Secure-aware Fine-tuning ICSE 2026

Fang Liu, Simiao Liu, Yinghao Zhu, Xiaoli Lian, Li Zhang

Identifying and addressing security issues during the early phase of the development lifecycle is critical for mitigating the long-term negative impacts on software systems. Code review serves as an effective practice that enables developers to check their teammates' code before integration into the codebase. To streamline the generation of review comments, various automated code review approaches have been proposed, where LLM-based methods have significantly advanced the capabilities of automated review generation. However, existing models primarily focus on general-purpose code review, their effectiveness in identifying and addressing security-related issues remains underexplored. Moreover, adapting existing code review approaches to target security issues faces substantial challenges, including data scarcity and inadequate evaluation metrics. To address these limitations, we propose SecureReviewer, a new approach designed for enhancing LLMs' ability to identify and resolve security-related issues during code review. Specifically, we first construct a dataset tailored for training and evaluating secure code review capabilities. Leveraging this dataset, we fine-tune LLMs to generate code review comments that can effectively identify security issues and provide fix suggestions with our proposed secure-aware fine-tuning strategy. To mitigate hallucination in LLMs and enhance the reliability of their outputs, we integrate the RAG technique, which grounds the generated comments in domain-specific security knowledge. Additionally, we introduce SecureBLEU, a new evaluation metric designed to assess the effectiveness of review comments in addressing security issues. Experimental results demonstrate that SecureReviewer outperforms state-of-the-art baselines in both security issue detection accuracy and the overall quality and practical utility of generated review comments.

comment: Accepted by ICSE 2026. Code and data: https://github.com/SIMIAO515/SecureReviewer

☆ 1+1>2: A Synergistic Sparse and Low-Rank Compression Method for Large Language Models EMNLP 2025

Zeliang Zong, Kai Zhang, Zheyang Li, Wenming Tan, Ye Ren, Yiyan Zhai, Jilin Hu

Large Language Models (LLMs) have demonstrated remarkable proficiency in language comprehension and generation; however, their widespread adoption is constrained by substantial bandwidth and computational demands. While pruning and low-rank approximation have each demonstrated promising performance individually, their synergy for LLMs remains underexplored. We introduce \underline{S}ynergistic \underline{S}parse and \underline{L}ow-Rank \underline{C}ompression (SSLC) methods for LLMs, which leverages the strengths of both techniques: low-rank approximation compresses the model by retaining its essential structure with minimal information loss, whereas sparse optimization eliminates non-essential weights, preserving those crucial for generalization. Based on theoretical analysis, we first formulate the low-rank approximation and sparse optimization as a unified problem and solve it by iterative optimization algorithm. Experiments on LLaMA and Qwen2.5 models (7B-70B) show that SSLC, without any additional training steps, consistently surpasses standalone methods, achieving state-of-the-arts results. Notably, SSLC compresses Qwen2.5 by 50\% with no performance drop and achieves at least 1.63$\times$ speedup, offering a practical solution for efficient LLM deployment.

comment: 15 pages, 6 figures, EMNLP 2025 findings

☆ Nexus: Execution-Grounded Multi-Agent Test Oracle Synthesis

Dong Huang, Mingzhe Du, Jie M. Zhang, Zheng Lin, Meng Luo, Qianru Zhang, See-Kiong Ng

Test oracle generation in non-regression testing is a longstanding challenge in software engineering, where the goal is to produce oracles that can accurately determine whether a function under test (FUT) behaves as intended for a given input. In this paper, we introduce Nexus, a novel multi-agent framework to address this challenge. Nexus generates test oracles by leveraging a diverse set of specialized agents that synthesize test oracles through a structured process of deliberation, validation, and iterative self-refinement. During the deliberation phase, a panel of four specialist agents, each embodying a distinct testing philosophy, collaboratively critiques and refines an initial set of test oracles. Then, in the validation phase, Nexus generates a plausible candidate implementation of the FUT and executes the proposed oracles against it in a secure sandbox. For any oracle that fails this execution-based check, Nexus activates an automated selfrefinement loop, using the specific runtime error to debug and correct the oracle before re-validation. Our extensive evaluation on seven diverse benchmarks demonstrates that Nexus consistently and substantially outperforms state-of-theart baselines. For instance, Nexus improves the test-level oracle accuracy on the LiveCodeBench from 46.30% to 57.73% for GPT-4.1-Mini. The improved accuracy also significantly enhances downstream tasks: the bug detection rate of GPT4.1-Mini generated test oracles on HumanEval increases from 90.91% to 95.45% for Nexus compared to baselines, and the success rate of automated program repair improves from 35.23% to 69.32%.

comment: Under Review

☆ OmniEduBench: A Comprehensive Chinese Benchmark for Evaluating Large Language Models in Education

Min Zhang, Hao Chen, Hao Chen, Wenqi Zhang, Didi Zhu, Xin Lin, Bo Jiang, Aimin Zhou, Fei Wu, Kun Kuang

With the rapid development of large language models (LLMs), various LLM-based works have been widely applied in educational fields. However, most existing LLMs and their benchmarks focus primarily on the knowledge dimension, largely neglecting the evaluation of cultivation capabilities that are essential for real-world educational scenarios. Additionally, current benchmarks are often limited to a single subject or question type, lacking sufficient diversity. This issue is particularly prominent within the Chinese context. To address this gap, we introduce OmniEduBench, a comprehensive Chinese educational benchmark. OmniEduBench consists of 24.602K high-quality question-answer pairs. The data is meticulously divided into two core dimensions: the knowledge dimension and the cultivation dimension, which contain 18.121K and 6.481K entries, respectively. Each dimension is further subdivided into 6 fine-grained categories, covering a total of 61 different subjects (41 in the knowledge and 20 in the cultivation). Furthermore, the dataset features a rich variety of question formats, including 11 common exam question types, providing a solid foundation for comprehensively evaluating LLMs' capabilities in education. Extensive experiments on 11 mainstream open-source and closed-source LLMs reveal a clear performance gap. In the knowledge dimension, only Gemini-2.5 Pro surpassed 60\% accuracy, while in the cultivation dimension, the best-performing model, QWQ, still trailed human intelligence by nearly 30\%. These results highlight the substantial room for improvement and underscore the challenges of applying LLMs in education.

☆ On the Role of Context for Discourse Relation Classification in Scientific Writing

Stephen Wan, Wei Liu, Michael Strube

With the increasing use of generative Artificial Intelligence (AI) methods to support science workflows, we are interested in the use of discourse-level information to find supporting evidence for AI generated scientific claims. A first step towards this objective is to examine the task of inferring discourse structure in scientific writing. In this work, we present a preliminary investigation of pretrained language model (PLM) and Large Language Model (LLM) approaches for Discourse Relation Classification (DRC), focusing on scientific publications, an under-studied genre for this task. We examine how context can help with the DRC task, with our experiments showing that context, as defined by discourse structure, is generally helpful. We also present an analysis of which scientific discourse relation types might benefit most from context.

comment: Accepted at Joint Sixth Workshop on Computational Approaches to Discourse, Context and Document-Level Inferences (CODI 2025) and Eighth Workshop on Computational Models of Reference, Anaphora and Coreference (CRAC 2025)

☆ The Geometry of Dialogue: Graphing Language Models to Reveal Synergistic Teams for Multi-Agent Collaboration

Kotaro Furuya, Yuichi Kitagawa

While a multi-agent approach based on large language models (LLMs) represents a promising strategy to surpass the capabilities of single models, its success is critically dependent on synergistic team composition. However, forming optimal teams is a significant challenge, as the inherent opacity of most models obscures the internal characteristics necessary for effective collaboration. In this paper, we propose an interaction-centric framework for automatic team composition that does not require any prior knowledge including their internal architectures, training data, or task performances. Our method constructs a "language model graph" that maps relationships between models from the semantic coherence of pairwise conversations, and then applies community detection to identify synergistic model clusters. Our experiments with diverse LLMs demonstrate that the proposed method discovers functionally coherent groups that reflect their latent specializations. Priming conversations with specific topics identified synergistic teams which outperform random baselines on downstream benchmarks and achieve comparable accuracy to that of manually-curated teams based on known model specializations. Our findings provide a new basis for the automated design of collaborative multi-agent LLM teams.

☆ MisSynth: Improving MISSCI Logical Fallacies Classification with Synthetic Data

Mykhailo Poliakov, Nadiya Shvai

Health-related misinformation is very prevalent and potentially harmful. It is difficult to identify, especially when claims distort or misinterpret scientific findings. We investigate the impact of synthetic data generation and lightweight fine-tuning techniques on the ability of large language models (LLMs) to recognize fallacious arguments using the MISSCI dataset and framework. In this work, we propose MisSynth, a pipeline that applies retrieval-augmented generation (RAG) to produce synthetic fallacy samples, which are then used to fine-tune an LLM model. Our results show substantial accuracy gains with fine-tuned models compared to vanilla baselines. For instance, the LLaMA 3.1 8B fine-tuned model achieved an over 35% F1-score absolute improvement on the MISSCI test split over its vanilla baseline. We demonstrate that introducing synthetic fallacy data to augment limited annotated resources can significantly enhance zero-shot LLM classification performance on real-world scientific misinformation tasks, even with limited computational resources. The code and synthetic dataset are available on https://github.com/mxpoliakov/MisSynth.

☆ From Amateur to Master: Infusing Knowledge into LLMs via Automated Curriculum Learning

Nishit Neema, Srinjoy Mukherjee, Sapan Shah, Gokul Ramakrishnan, Ganesh Venkatesh

Large Language Models (LLMs) excel at general tasks but underperform in specialized domains like economics and psychology, which require deep, principled understanding. To address this, we introduce ACER (Automated Curriculum-Enhanced Regimen) that transforms generalist models into domain experts without sacrificing their broad capabilities. ACER first synthesizes a comprehensive, textbook-style curriculum by generating a table of contents for a subject and then creating question-answer (QA) pairs guided by Bloom's taxonomy. This ensures systematic topic coverage and progressively increasing difficulty. The resulting synthetic corpus is used for continual pretraining with an interleaved curriculum schedule, aligning learning across both content and cognitive dimensions. Experiments with Llama 3.2 (1B and 3B) show significant gains in specialized MMLU subsets. In challenging domains like microeconomics, where baselines struggle, ACER boosts accuracy by 5 percentage points. Across all target domains, we observe a consistent macro-average improvement of 3 percentage points. Notably, ACER not only prevents catastrophic forgetting but also facilitates positive cross-domain knowledge transfer, improving performance on non-target domains by 0.7 points. Beyond MMLU, ACER enhances performance on knowledge-intensive benchmarks like ARC and GPQA by over 2 absolute points, while maintaining stable performance on general reasoning tasks. Our results demonstrate that ACER offers a scalable and effective recipe for closing critical domain gaps in LLMs.

☆ SCRIBE: Structured Chain Reasoning for Interactive Behaviour Explanations using Tool Calling

Fares Fawzi, Vinitra Swamy, Dominik Glandorf, Tanya Nazaretsky, Tanja Käser

Language models can be used to provide interactive, personalized student feedback in educational settings. However, real-world deployment faces three key challenges: privacy concerns, limited computational resources, and the need for pedagogically valid responses. These constraints require small, open-source models that can run locally and reliably ground their outputs in correct information. We introduce SCRIBE, a framework for multi-hop, tool-augmented reasoning designed to generate valid responses to student questions about feedback reports. SCRIBE combines domain-specific tools with a self-reflective inference pipeline that supports iterative reasoning, tool use, and error recovery. We distil these capabilities into 3B and 8B models via two-stage LoRA fine-tuning on synthetic GPT-4o-generated data. Evaluation with a human-aligned GPT-Judge and a user study with 108 students shows that 8B-SCRIBE models achieve comparable or superior quality to much larger models in key dimensions such as relevance and actionability, while being perceived on par with GPT-4o and Llama-3.3 70B by students. These findings demonstrate the viability of SCRIBE for low-resource, privacy-sensitive educational applications.

☆ Can Agent Conquer Web? Exploring the Frontiers of ChatGPT Atlas Agent in Web Games

Jingran Zhang, Ning Li, Justin Cui

OpenAI's ChatGPT Atlas introduces new capabilities for web interaction, enabling the model to analyze webpages, process user intents, and execute cursor and keyboard inputs directly within the browser. While its capacity for information retrieval tasks has been demonstrated, its performance in dynamic, interactive environments remains less explored. In this study, we conduct an early evaluation of Atlas's web interaction capabilities using browser-based games as test scenarios, including Google's T-Rex Runner, Sudoku, Flappy Bird, and Stein.world. We employ in-game performance scores as quantitative metrics to assess performance across different task types. Our results show that Atlas performs strongly in logical reasoning tasks like Sudoku, completing puzzles significantly faster than human baselines, but struggles substantially in real-time games requiring precise timing and motor control, often failing to progress beyond initial obstacles. These findings suggest that while Atlas demonstrates capable analytical processing, there remain notable limitations in dynamic web environments requiring real-time interaction. The website of our project can be found at https://atlas-game-eval.github.io.

☆ Unravelling the Mechanisms of Manipulating Numbers in Language Models

Michal Štefánik, Timothee Mickus, Marek Kadlčík, Bertram Højer, Michal Spiegel, Raúl Vázquez, Aman Sinha, Josef Kuchař, Philipp Mondorf

Recent work has shown that different large language models (LLMs) converge to similar and accurate input embedding representations for numbers. These findings conflict with the documented propensity of LLMs to produce erroneous outputs when dealing with numeric information. In this work, we aim to explain this conflict by exploring how language models manipulate numbers and quantify the lower bounds of accuracy of these mechanisms. We find that despite surfacing errors, different language models learn interchangeable representations of numbers that are systematic, highly accurate and universal across their hidden states and the types of input contexts. This allows us to create universal probes for each LLM and to trace information -- including the causes of output errors -- to specific layers. Our results lay a fundamental understanding of how pre-trained LLMs manipulate numbers and outline the potential of more accurate probing techniques in addressed refinements of LLMs' architectures.

☆ Do LLMs Signal When They're Right? Evidence from Neuron Agreement

Kang Chen, Yaoning Wang, Kai Xiong, Zhuoka Feng, Wenhe Sun, Haotian Chen, Yixin Cao

Large language models (LLMs) commonly boost reasoning via sample-evaluate-ensemble decoders, achieving label free gains without ground truth. However, prevailing strategies score candidates using only external outputs such as token probabilities, entropies, or self evaluations, and these signals can be poorly calibrated after post training. We instead analyze internal behavior based on neuron activations and uncover three findings: (1) external signals are low dimensional projections of richer internal dynamics; (2) correct responses activate substantially fewer unique neurons than incorrect ones throughout generation; and (3) activations from correct responses exhibit stronger cross sample agreement, whereas incorrect ones diverge. Motivated by these observations, we propose Neuron Agreement Decoding (NAD), an unsupervised best-of-N method that selects candidates using activation sparsity and cross sample neuron agreement, operating solely on internal signals and without requiring comparable textual outputs. NAD enables early correctness prediction within the first 32 generated tokens and supports aggressive early stopping. Across math and science benchmarks with verifiable answers, NAD matches majority voting; on open ended coding benchmarks where majority voting is inapplicable, NAD consistently outperforms Avg@64. By pruning unpromising trajectories early, NAD reduces token usage by 99% with minimal loss in generation quality, showing that internal signals provide reliable, scalable, and efficient guidance for label free ensemble decoding.

☆ PVMark: Enabling Public Verifiability for LLM Watermarking Schemes

Haohua Duan, Liyao Xiang, Xin Zhang

Watermarking schemes for large language models (LLMs) have been proposed to identify the source of the generated text, mitigating the potential threats emerged from model theft. However, current watermarking solutions hardly resolve the trust issue: the non-public watermark detection cannot prove itself faithfully conducting the detection. We observe that it is attributed to the secret key mostly used in the watermark detection -- it cannot be public, or the adversary may launch removal attacks provided the key; nor can it be private, or the watermarking detection is opaque to the public. To resolve the dilemma, we propose PVMark, a plugin based on zero-knowledge proof (ZKP), enabling the watermark detection process to be publicly verifiable by third parties without disclosing any secret key. PVMark hinges upon the proof of `correct execution' of watermark detection on which a set of ZKP constraints are built, including mapping, random number generation, comparison, and summation. We implement multiple variants of PVMark in Python, Rust and Circom, covering combinations of three watermarking schemes, three hash functions, and four ZKP protocols, to show our approach effectively works under a variety of circumstances. By experimental results, PVMark efficiently enables public verifiability on the state-of-the-art LLM watermarking schemes yet without compromising the watermarking performance, promising to be deployed in practice.

comment: This work has been submitted to the IEEE for possible publication

☆ Distilling Multilingual Vision-Language Models: When Smaller Models Stay Multilingual

Sukrit Sriratanawilai, Jhayahgrit Thongwat, Romrawin Chumpu, Patomporn Payoungkhamdee, Sarana Nutanong, Peerat Limkonchotiwat

Vision-language models (VLMs) exhibit uneven performance across languages, a problem that is often exacerbated when the model size is reduced. While Knowledge distillation (KD) demonstrates promising results in transferring knowledge from larger to smaller VLMs, applying KD in multilingualism is an underexplored area. This paper presents a controlled empirical study of KD behavior across five distillation approaches, isolating their effects on cross-lingual representation consistency and downstream performance stability under model compression. We study five distillation formulations across CLIP and SigLIP2, and evaluate them on in-domain retrieval and out-of-domain visual QA. We find that some configurations preserve or even improve multilingual retrieval robustness despite halving model size, but others fail to maintain cross-task stability, exposing design-sensitive trade-offs that aggregate accuracy alone does not reveal.

comment: Work in progress

☆ Language Models Are Borrowing-Blind: A Multilingual Evaluation of Loanword Identification across 10 Languages

Mérilin Sousa Silva, Sina Ahmadi

Throughout language history, words are borrowed from one language to another and gradually become integrated into the recipient's lexicon. Speakers can often differentiate these loanwords from native vocabulary, particularly in bilingual communities where a dominant language continuously imposes lexical items on a minority language. This paper investigates whether pretrained language models, including large language models, possess similar capabilities for loanword identification. We evaluate multiple models across 10 languages. Despite explicit instructions and contextual information, our results show that models perform poorly in distinguishing loanwords from native ones. These findings corroborate previous evidence that modern NLP systems exhibit a bias toward loanwords rather than native equivalents. Our work has implications for developing NLP tools for minority languages and supporting language preservation in communities under lexical pressure from dominant languages.

comment: Under review

☆ Pragmatic Theories Enhance Understanding of Implied Meanings in LLMs

Takuma Sato, Seiya Kawano, Koichiro Yoshino

The ability to accurately interpret implied meanings plays a crucial role in human communication and language use, and language models are also expected to possess this capability. This study demonstrates that providing language models with pragmatic theories as prompts is an effective in-context learning approach for tasks to understand implied meanings. Specifically, we propose an approach in which an overview of pragmatic theories, such as Gricean pragmatics and Relevance Theory, is presented as a prompt to the language model, guiding it through a step-by-step reasoning process to derive a final interpretation. Experimental results showed that, compared to the baseline, which prompts intermediate reasoning without presenting pragmatic theories (0-shot Chain-of-Thought), our methods enabled language models to achieve up to 9.6\% higher scores on pragmatic reasoning tasks. Furthermore, we show that even without explaining the details of pragmatic theories, merely mentioning their names in the prompt leads to a certain performance improvement (around 1-3%) in larger models compared to the baseline.

☆ Which Way Does Time Flow? A Psychophysics-Grounded Evaluation for Vision-Language Models

Shiho Matta, Lis Kanashiro Pereira, Peitao Han, Fei Cheng, Shigeru Kitazawa

Modern vision-language models (VLMs) excel at many multimodal tasks, yet their grasp of temporal information in video remains weak and, crucially, under-evaluated. We probe this gap with a deceptively simple but revealing challenge: judging the arrow of time (AoT)-whether a short clip is played forward or backward. We introduce AoT-PsyPhyBENCH, a psychophysically validated benchmark that tests whether VLMs can infer temporal direction in natural videos using the same stimuli and behavioral baselines established for humans. Our comprehensive evaluation of open-weight and proprietary, reasoning and non-reasoning VLMs reveals that most models perform near chance, and even the best lag far behind human accuracy on physically irreversible processes (e.g., free fall, diffusion/explosion) and causal manual actions (division/addition) that humans recognize almost instantly. These results highlight a fundamental gap in current multimodal systems: while they capture rich visual-semantic correlations, they lack the inductive biases required for temporal continuity and causal understanding. We release the code and data for AoT-PsyPhyBENCH to encourage further progress in the physical and temporal reasoning capabilities of VLMs.

comment: 10 pages

☆ Towards Global Retrieval Augmented Generation: A Benchmark for Corpus-Level Reasoning

Qi Luo, Xiaonan Li, Tingshuo Fan, Xinchi Chen, Xipeng Qiu

Retrieval-augmented generation (RAG) has emerged as a leading approach to reducing hallucinations in large language models (LLMs). Current RAG evaluation benchmarks primarily focus on what we call local RAG: retrieving relevant chunks from a small subset of documents to answer queries that require only localized understanding within specific text chunks. However, many real-world applications require a fundamentally different capability -- global RAG -- which involves aggregating and analyzing information across entire document collections to derive corpus-level insights (for example, "What are the top 10 most cited papers in 2023?"). In this paper, we introduce GlobalQA -- the first benchmark specifically designed to evaluate global RAG capabilities, covering four core task types: counting, extremum queries, sorting, and top-k extraction. Through systematic evaluation across different models and baselines, we find that existing RAG methods perform poorly on global tasks, with the strongest baseline achieving only 1.51 F1 score. To address these challenges, we propose GlobalRAG, a multi-tool collaborative framework that preserves structural coherence through chunk-level retrieval, incorporates LLM-driven intelligent filters to eliminate noisy documents, and integrates aggregation modules for precise symbolic computation. On the Qwen2.5-14B model, GlobalRAG achieves 6.63 F1 compared to the strongest baseline's 1.51 F1, validating the effectiveness of our method.

☆ What's In My Human Feedback? Learning Interpretable Descriptions of Preference Data

Rajiv Movva, Smitha Milli, Sewon Min, Emma Pierson

Human feedback can alter language models in unpredictable and undesirable ways, as practitioners lack a clear understanding of what feedback data encodes. While prior work studies preferences over certain attributes (e.g., length or sycophancy), automatically extracting relevant features without pre-specifying hypotheses remains challenging. We introduce What's In My Human Feedback? (WIMHF), a method to explain feedback data using sparse autoencoders. WIMHF characterizes both (1) the preferences a dataset is capable of measuring and (2) the preferences that the annotators actually express. Across 7 datasets, WIMHF identifies a small number of human-interpretable features that account for the majority of the preference prediction signal achieved by black-box models. These features reveal a wide diversity in what humans prefer, and the role of dataset-level context: for example, users on Reddit prefer informality and jokes, while annotators in HH-RLHF and PRISM disprefer them. WIMHF also surfaces potentially unsafe preferences, such as that LMArena users tend to vote against refusals, often in favor of toxic content. The learned features enable effective data curation: re-labeling the harmful examples in Arena yields large safety gains (+37%) with no cost to general performance. They also allow fine-grained personalization: on the Community Alignment dataset, we learn annotator-specific weights over subjective features that improve preference prediction. WIMHF provides a human-centered analysis method for practitioners to better understand and use preference data.

comment: Code: https://github.com/rmovva/wimhf

☆ Don't Let It Fade: Preserving Edits in Diffusion Language Models via Token Timestep Allocation NeurIPS 2025

Woojin Kim, Jaeyoung Do

While diffusion language models (DLMs) enable fine-grained refinement, their practical controllability remains fragile. We identify and formally characterize a central failure mode called update forgetting, in which uniform and context agnostic updates induce token level fluctuations across timesteps, erasing earlier semantic edits and disrupting the cumulative refinement process, thereby degrading fluency and coherence. As this failure originates in uniform and context agnostic updates, effective control demands explicit token ordering. We propose Token Timestep Allocation (TTA), which realizes soft and semantic token ordering via per token timestep schedules: critical tokens are frozen early, while uncertain tokens receive continued refinement. This timestep based ordering can be instantiated as either a fixed policy or an adaptive policy driven by task signals, thereby supporting a broad spectrum of refinement strategies. Because it operates purely at inference time, it applies uniformly across various DLMs and naturally extends to diverse supervision sources. Empirically, TTA improves controllability and fluency: on sentiment control, it yields more than 20 percent higher accuracy and nearly halves perplexity using less than one fifth the steps; in detoxification, it lowers maximum toxicity (12.2 versus 14.5) and perplexity (26.0 versus 32.0). Together, these results demonstrate that softened ordering via timestep allocation is the critical lever for mitigating update forgetting and achieving stable and controllable diffusion text generation.

comment: Accepted in NeurIPS 2025

☆ RCScore: Quantifying Response Consistency in Large Language Models

Dongjun Jang, Youngchae Ahn, Hyopil Shin

Current LLM evaluations often rely on a single instruction template, overlooking models' sensitivity to instruction style-a critical aspect for real-world deployments. We present RCScore, a multi-dimensional framework quantifying how instruction formulation affects model responses. By systematically transforming benchmark problems into multiple instruction styles, RCScore reveals performance variations undetected by conventional metrics. Our experiments across ten LLMs on four reasoning benchmarks demonstrate that instruction style can shift accuracy by up to 16.7% points. We introduce Cross-Response Similarity (CRS), a method applying RCScore metrics to measure stylistic self-consistency, and establish its strong correlation with task accuracy, suggesting consistency as a valuable proxy for model reliability. Additional findings show that deterministic decoding produces more stylistically stable outputs, and model scale correlates positively with cross-style consistency. RCScore offers a principled approach to assess instruction robustness.

☆ SP-MCQA: Evaluating Intelligibility of TTS Beyond the Word Level

Hitomi Jin Ling Tee, Chaoren Wang, Zijie Zhang, Zhizheng Wu

The evaluation of intelligibility for TTS has reached a bottleneck, as existing assessments heavily rely on word-by-word accuracy metrics such as WER, which fail to capture the complexity of real-world speech or reflect human comprehension needs. To address this, we propose Spoken-Passage Multiple-Choice Question Answering, a novel subjective approach evaluating the accuracy of key information in synthesized speech, and release SP-MCQA-Eval, an 8.76-hour news-style benchmark dataset for SP-MCQA evaluation. Our experiments reveal that low WER does not necessarily guarantee high key-information accuracy, exposing a gap between traditional metrics and practical intelligibility. SP-MCQA shows that even state-of-the-art (SOTA) models still lack robust text normalization and phonetic accuracy. This work underscores the urgent need for high-level, more life-like evaluation criteria now that many systems already excel at WER yet may fall short on real-world intelligibility.

☆ Similarity-Distance-Magnitude Language Models

Allen Schmaltz

We introduce Similarity-Distance-Magnitude (SDM) language models (LMs), which are sequence prediction models fine-tuned to maximize the proportion of generations in the well-calibrated, high-probability region partitioned by a final-layer SDM activation layer used for binary classification of instruction-following. We demonstrate that existing pre-trained decoder-only Transformer LMs can be readily converted into SDM LMs via supervised fine-tuning, using the final-layer SDM activation layer during training to estimate a change-of-base for a supervised next-token loss over a contrastive input encoding scheme, with additional hard negative examples generated online during training. This results in reduced abstentions (i.e., improved statistical efficiency) compared to strong supervised baselines.

comment: 8 pages, 5 tables

☆ MossNet: Mixture of State-Space Experts is a Multi-Head Attention

Shikhar Tuli, James Seale Smith, Haris Jeelani, Chi-Heng Lin, Abhishek Patel, Vasili Ramanishka, Yen-Chang Hsu, Hongxia Jin

Large language models (LLMs) have significantly advanced generative applications in natural language processing (NLP). Recent trends in model architectures revolve around efficient variants of transformers or state-space/gated-recurrent models (SSMs, GRMs). However, prevailing SSM/GRM-based methods often emulate only a single attention head, potentially limiting their expressiveness. In this work, we propose MossNet, a novel mixture-of-state-space-experts architecture that emulates a linear multi-head attention (MHA). MossNet leverages a mixture-of-experts (MoE) implementation not only in channel-mixing multi-layered perceptron (MLP) blocks but also in the time-mixing SSM kernels to realize multiple "attention heads." Extensive experiments on language modeling and downstream evaluations show that MossNet outperforms both transformer- and SSM-based architectures of similar model size and data budgets. Larger variants of MossNet, trained on trillions of tokens, further confirm its scalability and superior performance. In addition, real-device profiling on a Samsung Galaxy S24 Ultra and an Nvidia A100 GPU demonstrate favorable runtime speed and resource usage compared to similarly sized baselines. Our results suggest that MossNet is a compelling new direction for efficient, high-performing recurrent LLM architectures.

☆ One Model to Critique Them All: Rewarding Agentic Tool-Use via Efficient Reasoning

Renhao Li, Jianhong Tu, Yang Su, Hamid Alinejad-Rokny, Derek F. Wong, Junyang Lin, Min Yang

Reward models (RMs) play a critical role in aligning large language models (LLMs) with human preferences. Yet in the domain of tool learning, the lack of RMs specifically designed for function-calling tasks has limited progress toward more capable agentic AI. We introduce ToolRM, a family of lightweight generative RMs tailored for general tool-use scenarios. To build these models, we propose a novel pipeline that constructs pairwise preference data using rule-based scoring and multidimensional sampling. This yields ToolPref-Pairwise-30K, a diverse, balanced, and challenging dataset of critique tasks that supports reinforcement learning with verifiable feedback. To evaluate tool-use RMs, we also introduce TRBench$_{BFCL}$, a benchmark built on the agentic evaluation suite BFCL. Trained on our constructed data, models from the Qwen3-4B/8B series achieve up to 14.28% higher accuracy, substantially outperforming frontier models such as Claude 4 and OpenAI o3 in pairwise reward judgments. Beyond training objectives, ToolRM generalizes to broader critique tasks, including Best-of-N sampling and self-correction. Experiments on ACEBench highlight its effectiveness and efficiency, enabling inference-time scaling and reducing output token usage by over 66%. We release data and model checkpoints to facilitate future research.

☆ Reasoning Curriculum: Bootstrapping Broad LLM Reasoning from Math

Bo Pang, Deqian Kong, Silvio Savarese, Caiming Xiong, Yingbo Zhou

Reinforcement learning (RL) can elicit strong reasoning in large language models (LLMs), yet most open efforts focus on math and code. We propose Reasoning Curriculum, a simple two-stage curriculum that first elicits reasoning skills in pretraining-aligned domains such as math, then adapts and refines these skills across other domains via joint RL. Stage 1 performs a brief cold start and then math-only RL with verifiable rewards to develop reasoning skills. Stage 2 runs joint RL on mixed-domain data to transfer and consolidate these skills. The curriculum is minimal and backbone-agnostic, requiring no specialized reward models beyond standard verifiability checks. Evaluated on Qwen3-4B and Llama-3.1-8B over a multi-domain suite, reasoning curriculum yields consistent gains. Ablations and a cognitive-skill analysis indicate that both stages are necessary and that math-first elicitation increases cognitive behaviors important for solving complex problems. Reasoning Curriculum provides a compact, easy-to-adopt recipe for general reasoning.

comment: 9 pages

☆ On the Influence of Discourse Relations in Persuasive Texts

Nawar Turk, Sevag Kaspar, Leila Kosseim

This paper investigates the relationship between Persuasion Techniques (PTs) and Discourse Relations (DRs) by leveraging Large Language Models (LLMs) and prompt engineering. Since no dataset annotated with both PTs and DRs exists, we took the SemEval 2023 Task 3 dataset labelled with 19 PTs as a starting point and developed LLM-based classifiers to label each instance of the dataset with one of the 22 PDTB 3.0 level-2 DRs. In total, four LLMs were evaluated using 10 different prompts, resulting in 40 unique DR classifiers. Ensemble models using different majority-pooling strategies were used to create 5 silver datasets of instances labelled with both persuasion techniques and level-2 PDTB senses. The silver dataset sizes vary from 1,281 instances to 204 instances, depending on the majority pooling technique used. Statistical analysis of these silver datasets shows that six discourse relations (namely Cause, Purpose, Contrast, Cause+Belief, Concession, and Condition) play a crucial role in persuasive texts, especially in the use of Loaded Language, Exaggeration/Minimisation, Repetition and to cast Doubt. This insight can contribute to detecting online propaganda and misinformation, as well as to our general understanding of effective communication.

comment: Published in Proceedings of the 38th Canadian Conference on Artificial Intelligence CanAI 2025 Calgary Alberta May 26-27 2025. 5 figures 7 tables

☆ Reasoning Path Divergence: A New Metric and Curation Strategy to Unlock LLM Diverse Thinking

Feng Ju, Zeyu Qin, Rui Min, Zhitao He, Lingpeng Kong, Yi R. Fung

While Test-Time Scaling (TTS) has proven effective in improving the reasoning ability of large language models (LLMs), low diversity in model outputs often becomes a bottleneck; this is partly caused by the common "one problem, one solution" (1P1S) training practice, which provides a single canonical answer and can push models toward a narrow set of reasoning paths. To address this, we propose a "one problem, multiple solutions" (1PNS) training paradigm that exposes the model to a variety of valid reasoning trajectories and thus increases inference diversity. A core challenge for 1PNS is reliably measuring semantic differences between multi-step chains of thought, so we introduce Reasoning Path Divergence (RPD), a step-level metric that aligns and scores Long Chain-of-Thought solutions to capture differences in intermediate reasoning. Using RPD, we curate maximally diverse solution sets per problem and fine-tune Qwen3-4B-Base. Experiments show that RPD-selected training yields more varied outputs and higher pass@k, with an average +2.80% gain in pass@16 over a strong 1P1S baseline and a +4.99% gain on AIME24, demonstrating that 1PNS further amplifies the effectiveness of TTS. Our code is available at https://github.com/fengjujf/Reasoning-Path-Divergence .

☆ QCoder Benchmark: Bridging Language Generation and Quantum Hardware through Simulator-Based Feedback

Taku Mikuriya, Tatsuya Ishigaki, Masayuki Kawarada, Shunya Minami, Tadashi Kadowaki, Yohichi Suzuki, Soshun Naito, Shunya Takata, Takumi Kato, Tamotsu Basseda, Reo Yamada, Hiroya Takamura

Large language models (LLMs) have increasingly been applied to automatic programming code generation. This task can be viewed as a language generation task that bridges natural language, human knowledge, and programming logic. However, it remains underexplored in domains that require interaction with hardware devices, such as quantum programming, where human coders write Python code that is executed on a quantum computer. To address this gap, we introduce QCoder Benchmark, an evaluation framework that assesses LLMs on quantum programming with feedback from simulated hardware devices. Our benchmark offers two key features. First, it supports evaluation using a quantum simulator environment beyond conventional Python execution, allowing feedback of domain-specific metrics such as circuit depth, execution time, and error classification, which can be used to guide better generation. Second, it incorporates human-written code submissions collected from real programming contests, enabling both quantitative comparisons and qualitative analyses of LLM outputs against human-written codes. Our experiments reveal that even advanced models like GPT-4o achieve only around 18.97% accuracy, highlighting the difficulty of the benchmark. In contrast, reasoning-based models such as o3 reach up to 78% accuracy, outperforming averaged success rates of human-written codes (39.98%). We release the QCoder Benchmark dataset and public evaluation API to support further research.

☆ ORBIT -- Open Recommendation Benchmark for Reproducible Research with Hidden Tests NeurIPS 2025

Jingyuan He, Jiongnan Liu, Vishan Vishesh Oberoi, Bolin Wu, Mahima Jagadeesh Patel, Kangrui Mao, Chuning Shi, I-Ta Lee, Arnold Overwijk, Chenyan Xiong

Recommender systems are among the most impactful AI applications, interacting with billions of users every day, guiding them to relevant products, services, or information tailored to their preferences. However, the research and development of recommender systems are hindered by existing datasets that fail to capture realistic user behaviors and inconsistent evaluation settings that lead to ambiguous conclusions. This paper introduces the Open Recommendation Benchmark for Reproducible Research with HIdden Tests (ORBIT), a unified benchmark for consistent and realistic evaluation of recommendation models. ORBIT offers a standardized evaluation framework of public datasets with reproducible splits and transparent settings for its public leaderboard. Additionally, ORBIT introduces a new webpage recommendation task, ClueWeb-Reco, featuring web browsing sequences from 87 million public, high-quality webpages. ClueWeb-Reco is a synthetic dataset derived from real, user-consented, and privacy-guaranteed browsing data. It aligns with modern recommendation scenarios and is reserved as the hidden test part of our leaderboard to challenge recommendation models' generalization ability. ORBIT measures 12 representative recommendation models on its public benchmark and introduces a prompted LLM baseline on the ClueWeb-Reco hidden test. Our benchmark results reflect general improvements of recommender systems on the public datasets, with variable individual performances. The results on the hidden test reveal the limitations of existing approaches in large-scale webpage recommendation and highlight the potential for improvements with LLM integrations. ORBIT benchmark, leaderboard, and codebase are available at https://www.open-reco-bench.ai.

comment: Accepted to NeurIPS 2025 Datasets & Benchmarks track

☆ Do Students Debias Like Teachers? On the Distillability of Bias Mitigation Methods

Jiali Cheng, Chirag Agarwal, Hadi Amiri

Knowledge distillation (KD) is an effective method for model compression and transferring knowledge between models. However, its effect on model's robustness against spurious correlations that degrade performance on out-of-distribution data remains underexplored. This study investigates the effect of knowledge distillation on the transferability of ``debiasing'' capabilities from teacher models to student models on natural language inference (NLI) and image classification tasks. Through extensive experiments, we illustrate several key findings: (i) overall the debiasing capability of a model is undermined post-KD; (ii) training a debiased model does not benefit from injecting teacher knowledge; (iii) although the overall robustness of a model may remain stable post-distillation, significant variations can occur across different types of biases; and (iv) we pin-point the internal attention pattern and circuit that causes the distinct behavior post-KD. Given the above findings, we propose three effective solutions to improve the distillability of debiasing methods: developing high quality data for augmentation, implementing iterative knowledge distillation, and initializing student models with weights obtained from teacher models. To the best of our knowledge, this is the first study on the effect of KD on debiasing and its interenal mechanism at scale. Our findings provide understandings on how KD works and how to design better debiasing methods.

☆ SIRAJ: Diverse and Efficient Red-Teaming for LLM Agents via Distilled Structured Reasoning

Kaiwen Zhou, Ahmed Elgohary, A S M Iftekhar, Amin Saied

The ability of LLM agents to plan and invoke tools exposes them to new safety risks, making a comprehensive red-teaming system crucial for discovering vulnerabilities and ensuring their safe deployment. We present SIRAJ: a generic red-teaming framework for arbitrary black-box LLM agents. We employ a dynamic two-step process that starts with an agent definition and generates diverse seed test cases that cover various risk outcomes, tool-use trajectories, and risk sources. Then, it iteratively constructs and refines model-based adversarial attacks based on the execution trajectories of former attempts. To optimize the red-teaming cost, we present a model distillation approach that leverages structured forms of a teacher model's reasoning to train smaller models that are equally effective. Across diverse evaluation agent settings, our seed test case generation approach yields 2 -- 2.5x boost to the coverage of risk outcomes and tool-calling trajectories. Our distilled 8B red-teamer model improves attack success rate by 100%, surpassing the 671B Deepseek-R1 model. Our ablations and analyses validate the effectiveness of the iterative framework, structured reasoning, and the generalization of our red-teamer models.

☆ Artificial Intelligence-Enabled Analysis of Radiology Reports: Epidemiology and Consequences of Incidental Thyroid Findings

Felipe Larios, Mariana Borras-Osorio, Yuqi Wu, Ana Gabriela Claros, David Toro-Tobon, Esteban Cabezas, Ricardo Loor-Torres, Maria Mateo Chavez, Kerly Guevara Maldonado, Luis Vilatuna Andrango, Maria Lizarazo Jimenez, Ivan Mateo Alzamora, Misk Al Zahidy, Marcelo Montero, Ana Cristina Proano, Cristian Soto Jacome, Jungwei W. Fan, Oscar J. Ponce-Ponte, Megan E. Branda, Naykky Singh Ospina, Juan P. Brito

Importance Incidental thyroid findings (ITFs) are increasingly detected on imaging performed for non-thyroid indications. Their prevalence, features, and clinical consequences remain undefined. Objective To develop, validate, and deploy a natural language processing (NLP) pipeline to identify ITFs in radiology reports and assess their prevalence, features, and clinical outcomes. Design, Setting, and Participants Retrospective cohort of adults without prior thyroid disease undergoing thyroid-capturing imaging at Mayo Clinic sites from July 1, 2017, to September 30, 2023. A transformer-based NLP pipeline identified ITFs and extracted nodule characteristics from image reports from multiple modalities and body regions. Main Outcomes and Measures Prevalence of ITFs, downstream thyroid ultrasound, biopsy, thyroidectomy, and thyroid cancer diagnosis. Logistic regression identified demographic and imaging-related factors. Results Among 115,683 patients (mean age, 56.8 [SD 17.2] years; 52.9% women), 9,077 (7.8%) had an ITF, of which 92.9% were nodules. ITFs were more likely in women, older adults, those with higher BMI, and when imaging was ordered by oncology or internal medicine. Compared with chest CT, ITFs were more likely via neck CT, PET, and nuclear medicine scans. Nodule characteristics were poorly documented, with size reported in 44% and other features in fewer than 15% (e.g. calcifications). Compared with patients without ITFs, those with ITFs had higher odds of thyroid nodule diagnosis, biopsy, thyroidectomy and thyroid cancer diagnosis. Most cancers were papillary, and larger when detected after ITFs vs no ITF. Conclusions ITFs were common and strongly associated with cascades leading to the detection of small, low-risk cancers. These findings underscore the role of ITFs in thyroid cancer overdiagnosis and the need for standardized reporting and more selective follow-up.

♻ ☆ TinyTim: A Family of Language Models for Divergent Generation NeurIPS

Christopher J. Agostino

In the search for artificial general intelligence, model development and training has focused primarily on vast datasets of known problems and their accepted solutions. This process necessarily produces convergent systems which are fundamentally incapable of the conceptual reframing that is required for genuine creative breakthroughs. Inspired by the divergent cognitive processes that allow humans to make such creative leaps, our work introduces a family of language models, TinyTim, to serve as sources of divergent generation within broader systems. These models have been created by fine-tuning on the anti-parsimonious text of James Joyce's `Finnegans Wake'. Quantitative analysis of both an unsupervised fine-tuned model (TinyTim-V1) and a new instruction-tuned variant (TinyTim-V2) demonstrates a profound capacity for lexical invention; the foundational V1 model exhibits a Yule's K score for lexical richness over twenty times greater than that of convergent baselines. This trait is a stable property of the family, as the instruction-tuned V2 maintains a statistically distinct profile and resists factual convergence, sacrificing benchmark performance to preserve its core generative style. This work establishes a methodology for engineering specialized divergent models that, when paired with convergent systems, can reframe problems and force breakthroughs beyond the reach of statistical optimization alone.

comment: 7 pages, 3 figures, accepted to NeurIPS Creative AI track, models available at https://hf.co/npc-worldwide/

♻ ☆ Completion $\neq$ Collaboration: Scaling Collaborative Effort with Agents

Shannon Zejiang Shen, Valerie Chen, Ken Gu, Alexis Ross, Zixian Ma, Jillian Ross, Alex Gu, Chenglei Si, Wayne Chi, Andi Peng, Jocelyn J Shen, Ameet Talwalkar, Tongshuang Wu, David Sontag

Current evaluations of agents remain centered around one-shot task completion, failing to account for the inherently iterative and collaborative nature of many real-world problems, where human goals are often underspecified and evolve. We argue for a shift from building and assessing task completion agents to developing collaborative agents, assessed not only by the quality of their final outputs but by how well they engage with and enhance human effort throughout the problem-solving process. To support this shift, we introduce collaborative effort scaling, a framework that captures how an agent's utility grows with increasing user involvement. Through case studies and simulated evaluations, we show that state-of-the-art agents often underperform in multi-turn, real-world scenarios, revealing a missing ingredient in agent design: the ability to sustain engagement and scaffold user understanding. Collaborative effort scaling offers a lens for diagnosing agent behavior and guiding development toward more effective interactions.

comment: 22 pages, 5 figures, 3 tables

♻ ☆ Comparing human and LLM politeness strategies in free production EMNLP 2025

Haoran Zhao, Robert D. Hawkins

Polite speech poses a fundamental alignment challenge for large language models (LLMs). Humans deploy a rich repertoire of linguistic strategies to balance informational and social goals -- from positive approaches that build rapport (compliments, expressions of interest) to negative strategies that minimize imposition (hedging, indirectness). We investigate whether LLMs employ a similarly context-sensitive repertoire by comparing human and LLM responses in both constrained and open-ended production tasks. We find that larger models ($\ge$70B parameters) successfully replicate key preferences from the computational pragmatics literature, and human evaluators surprisingly prefer LLM-generated responses in open-ended contexts. However, further linguistic analyses reveal that models disproportionately rely on negative politeness strategies even in positive contexts, potentially leading to misinterpretations. While modern LLMs demonstrate an impressive handle on politeness strategies, these subtle differences raise important questions about pragmatic alignment in AI systems.

comment: 25 pages, 5 figures | EMNLP 2025 camera-ready version

♻ ☆ Quality Over Quantity? LLM-Based Curation for a Data-Efficient Audio-Video Foundation Model

Ali Vosoughi, Dimitra Emmanouilidou, Hannes Gamper

Integrating audio and visual data for training multimodal foundational models remains a challenge. The Audio-Video Vector Alignment (AVVA) framework addresses this by considering AV scene alignment beyond mere temporal synchronization, and leveraging Large Language Models (LLMs) for data curation. AVVA implements a scoring mechanism for selecting aligned training data segments. It integrates Whisper, a speech-based foundation model, for audio and DINOv2 for video analysis in a dual-encoder structure with contrastive learning on AV pairs. Evaluations on AudioCaps, VALOR, and VGGSound demonstrate the effectiveness of the proposed model architecture and data curation approach. AVVA achieves a significant improvement in top-k accuracies for video-to-audio retrieval on all datasets compared to DenseAV, while using only 192 hrs of curated training data. Furthermore, an ablation study indicates that the data curation process effectively trades data quality for data quantity, yielding increases in top-k retrieval accuracies on AudioCaps, VALOR, and VGGSound, compared to training on the full spectrum of uncurated data.

comment: 5 pages, 5 figures, 2 tables. Accepted at EUSIPCO 2025

♻ ☆ Massive Supervised Fine-tuning Experiments Reveal How Data, Layer, and Training Factors Shape LLM Alignment Quality EMNLP 2025

Yuto Harada, Yusuke Yamauchi, Yusuke Oda, Yohei Oseki, Yusuke Miyao, Yu Takagi

Supervised fine-tuning (SFT) is a critical step in aligning large language models (LLMs) with human instructions and values, yet many aspects of SFT remain poorly understood. We trained a wide range of base models on a variety of datasets including code generation, mathematical reasoning, and general-domain tasks, resulting in 1,000+ SFT models under controlled conditions. We then identified the dataset properties that matter most and examined the layer-wise modifications introduced by SFT. Our findings reveal that some training-task synergies persist across all models while others vary substantially, emphasizing the importance of model-specific strategies. Moreover, we demonstrate that perplexity consistently predicts SFT effectiveness, often surpassing superficial similarity between the training data and the benchmark, and that mid-layer weight changes correlate most strongly with performance gains. We release these 1,000+ SFT models and benchmark results to accelerate further research. All resources are available at https://github.com/llm-jp/massive-sft.

comment: Accepted to EMNLP 2025 (Main Conference). Models and evaluation results available at: https://github.com/llm-jp/massive-sft

♻ ☆ Enhancing Reasoning Skills in Small Persian Medical Language Models Can Outperform Large-Scale Data Training

Mehrdad Ghassabi, Sadra Hakim, Hamidreza Baradaran Kashani, Pedram Rostami

Enhancing reasoning capabilities in small language models is critical for specialized applications such as medical question answering, particularly in underrepresented languages like Persian. In this study, we employ Reinforcement Learning with AI Feedback (RLAIF) and Direct preference optimization (DPO) to improve the reasoning skills of a general-purpose Persian language model. To achieve this, we translated a multiple-choice medical question-answering dataset into Persian and used RLAIF to generate rejected-preferred answer pairs, which are essential for DPO training. By prompting both teacher and student models to produce Chain-of-Thought (CoT) reasoning responses, we compiled a dataset containing correct and incorrect reasoning trajectories. This dataset, comprising 2 million tokens in preferred answers and 2.5 million tokens in rejected ones, was used to train a baseline model, significantly enhancing its medical reasoning capabilities in Persian. Remarkably, the resulting model outperformed its predecessor, gaokerena-V, which was trained on approximately 57 million tokens, despite leveraging a much smaller dataset. These results highlight the efficiency and effectiveness of reasoning-focused training approaches in developing domain-specific language models with limited data availability.

comment: 7 pages, 5 figures

♻ ☆ Controlling Thinking Speed in Reasoning Models NeurIPS 2025

Zhengkai Lin, Zhihang Fu, Ze Chen, Chao Chen, Liang Xie, Wenxiao Wang, Deng Cai, Zheng Wang, Jieping Ye

Human cognition is theorized to operate in two modes: fast, intuitive System 1 thinking and slow, deliberate System 2 thinking. While current Large Reasoning Models (LRMs) excel at System 2 thinking, their inability to perform fast thinking leads to high computational overhead and latency. In this work, we enable LRMs to approximate human intelligence through dynamic thinking speed adjustment, optimizing accuracy-efficiency trade-offs. Our approach addresses two key questions: (1) how to control thinking speed in LRMs, and (2) when to adjust it for optimal performance. For the first question, we identify the steering vector that governs slow-fast thinking transitions in LRMs' representation space. Using this vector, we achieve the first representation editing-based test-time scaling effect, outperforming existing prompt-based scaling methods. For the second question, we apply real-time difficulty estimation to signal reasoning segments of varying complexity. Combining these techniques, we propose the first reasoning strategy that enables fast processing of easy steps and deeper analysis for complex reasoning. Without any training or additional cost, our plug-in module delivers an average +1.3% accuracy with -8.6% token usage across leading LRMs and advanced reasoning benchmarks. All of our algorithms are implemented based on vLLM and are expected to support broader applications and inspire future research.

comment: NeurIPS 2025 Spotlight

♻ ☆ RLBFF: Binary Flexible Feedback to bridge between Human Feedback & Verifiable Rewards

Zhilin Wang, Jiaqi Zeng, Olivier Delalleau, Ellie Evans, Daniel Egert, Hoo-Chang Shin, Felipe Soares, Yi Dong, Oleksii Kuchaiev

Reinforcement Learning with Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) are the main RL paradigms used in LLM post-training, each offering distinct advantages. However, RLHF struggles with interpretability and reward hacking because it relies on human judgments that usually lack explicit criteria, whereas RLVR is limited in scope by its focus on correctness-based verifiers. We propose Reinforcement Learning with Binary Flexible Feedback (RLBFF), which combines the versatility of human-driven preferences with the precision of rule-based verification, enabling reward models to capture nuanced aspects of response quality beyond mere correctness. RLBFF extracts principles that can be answered in a binary fashion (e.g. accuracy of information: yes, or code readability: no) from natural language feedback. Such principles can then be used to ground Reward Model training as an entailment task (response satisfies or does not satisfy an arbitrary principle). We show that Reward Models trained in this manner can outperform Bradley-Terry models when matched for data and achieve top performance on RM-Bench (86.2%) and JudgeBench (81.4%, #1 on leaderboard as of September 24, 2025). Additionally, users can specify principles of interest at inference time to customize the focus of our reward models, in contrast to Bradley-Terry models. Finally, we present a fully open source recipe (including data) to align Qwen3-32B using RLBFF and our Reward Model, to match or exceed the performance of o3-mini and DeepSeek R1 on general alignment benchmarks of MT-Bench, WildBench, and Arena Hard v2 (at <5% of the inference cost). Models: https://huggingface.co/collections/nvidia/reward-models-10-2025

comment: Added link to access models: https://huggingface.co/collections/nvidia/reward-models-10-2025

♻ ☆ CompoST: A Benchmark for Analyzing the Ability of LLMs To Compositionally Interpret Questions in a QALD Setting ISWC 2025

David Maria Schmidt, Raoul Schubert, Philipp Cimiano

Language interpretation is a compositional process, in which the meaning of more complex linguistic structures is inferred from the meaning of their parts. Large language models possess remarkable language interpretation capabilities and have been successfully applied to interpret questions by mapping them to SPARQL queries. An open question is how systematic this interpretation process is. Toward this question, in this paper, we propose a benchmark for investigating to what extent the abilities of LLMs to interpret questions are actually compositional. For this, we generate three datasets of varying difficulty based on graph patterns in DBpedia, relying on Lemon lexica for verbalization. Our datasets are created in a very controlled fashion in order to test the ability of LLMs to interpret structurally complex questions, given that they have seen the atomic building blocks. This allows us to evaluate to what degree LLMs are able to interpret complex questions for which they "understand" the atomic parts. We conduct experiments with models of different sizes using both various prompt and few-shot optimization techniques as well as fine-tuning. Our results show that performance in terms of macro $F_1$ degrades from $0.45$ over $0.26$ down to $0.09$ with increasing deviation from the samples optimized on. Even when all necessary information was provided to the model in the input, the $F_1$ scores do not exceed $0.57$ for the dataset of lowest complexity. We thus conclude that LLMs struggle to systematically and compositionally interpret questions and map them into SPARQL queries.

comment: Research Track, 24th International Semantic Web Conference (ISWC 2025), November 2-6, 2025, Nara, Japan

♻ ☆ Unveiling Unicode's Unseen Underpinnings in Undermining Authorship Attribution

Robert Dilworth

When using a public communication channel -- whether formal or informal, such as commenting or posting on social media -- end users have no expectation of privacy: they compose a message and broadcast it for the world to see. Even if an end user takes utmost precautions to anonymize their online presence -- using an alias or pseudonym; masking their IP address; spoofing their geolocation; concealing their operating system and user agent; deploying encryption; registering with a disposable phone number or email; disabling non-essential settings; revoking permissions; and blocking cookies and fingerprinting -- one obvious element still lingers: the message itself. Assuming they avoid lapses in judgment or accidental self-exposure, there should be little evidence to validate their actual identity, right? Wrong. The content of their message -- necessarily open for public consumption -- exposes an attack vector: stylometric analysis, or author profiling. In this paper, we dissect the technique of stylometry, discuss an antithetical counter-strategy in adversarial stylometry, and devise enhancements through Unicode steganography.

comment: 33 pages, 7 figures, 3 tables

♻ ☆ Detecting Early and Implicit Suicidal Ideation via Longitudinal and Information Environment Signals on Social Media

Soorya Ram Shimgekar, Ruining Zhao, Agam Goyal, Violeta J. Rodriguez, Paul A. Bloom, Hari Sundaram, Koustuv Saha

On social media, many individuals experiencing suicidal ideation (SI) do not disclose their distress explicitly. Instead, signs may surface indirectly through everyday posts or peer interactions. Detecting such implicit signals early is critical but remains challenging. We frame early and implicit SI as a forward-looking prediction task and develop a computational framework that models a user's information environment, consisting of both their longitudinal posting histories as well as the discourse of their socially proximal peers. We adopted a composite network centrality measure to identify top neighbors of a user, and temporally aligned the user's and neighbors' interactions -- integrating the multi-layered signals in a fine-tuned DeBERTa-v3 model. In a Reddit study of 1,000 (500 Case and 500 Control) users, our approach improves early and implicit SI detection by 15% over individual-only baselines. These findings highlight that peer interactions offer valuable predictive signals and carry broader implications for designing early detection systems that capture indirect as well as masked expressions of risk in online environments.

♻ ☆ LatentBreak: Jailbreaking Large Language Models through Latent Space Feedback

Raffaele Mura, Giorgio Piras, Kamilė Lukošiūtė, Maura Pintor, Amin Karbasi, Battista Biggio

Jailbreaks are adversarial attacks designed to bypass the built-in safety mechanisms of large language models. Automated jailbreaks typically optimize an adversarial suffix or adapt long prompt templates by forcing the model to generate the initial part of a restricted or harmful response. In this work, we show that existing jailbreak attacks that leverage such mechanisms to unlock the model response can be detected by a straightforward perplexity-based filtering on the input prompt. To overcome this issue, we propose LatentBreak, a white-box jailbreak attack that generates natural adversarial prompts with low perplexity capable of evading such defenses. LatentBreak substitutes words in the input prompt with semantically-equivalent ones, preserving the initial intent of the prompt, instead of adding high-perplexity adversarial suffixes or long templates. These words are chosen by minimizing the distance in the latent space between the representation of the adversarial prompt and that of harmless requests. Our extensive evaluation shows that LatentBreak leads to shorter and low-perplexity prompts, thus outperforming competing jailbreak algorithms against perplexity-based filters on multiple safety-aligned models.

♻ ☆ Are You There God? Lightweight Narrative Annotation of Christian Fiction with LMs

Rebecca M. M. Hicke, Brian W. Haggard, Mia Ferrante, Rayhan Khanna, David Mimno

In addition to its more widely studied cultural movements, American Evangelicalism has a well-developed but less externally visible literary side. Christian Fiction, however, has been little studied, and what scholarly attention there is has focused on the explosively popular Left Behind series. In this work, we use computational tools to provide both a broad topical overview of Christian Fiction as a genre and a more directed exploration of how its authors depict divine acts. Working with human annotators, we first developed a codebook for identifying "acts of God." We then adapted the codebook for use by a recent, lightweight LM with the assistance of a much larger model. The laptop-scale LM is largely capable of matching human annotations, even when the task is subtle and challenging. Using these annotations, we show that significant and meaningful differences exist between divine acts depicted by the Left Behind books and Christian Fiction more broadly.

comment: Accepted to CHR 2025

♻ ☆ Unstructured Evidence Attribution for Long Context Query Focused Summarization EMNLP 2025

Dustin Wright, Zain Muhammad Mujahid, Lu Wang, Isabelle Augenstein, David Jurgens

Large language models (LLMs) are capable of generating coherent summaries from very long contexts given a user query, and extracting and citing evidence spans helps improve the trustworthiness of these summaries. Whereas previous work has focused on evidence citation with fixed levels of granularity (e.g. sentence, paragraph, document, etc.), we propose to extract unstructured (i.e., spans of any length) evidence in order to acquire more relevant and consistent evidence than in the fixed granularity case. We show how existing systems struggle to copy and properly cite unstructured evidence, which also tends to be "lost-in-the-middle". To help models perform this task, we create the Summaries with Unstructured Evidence Text dataset (SUnsET), a synthetic dataset generated using a novel pipeline, which can be used as training supervision for unstructured evidence summarization. We demonstrate across 5 LLMs and 4 datasets spanning human written, synthetic, single, and multi-document settings that LLMs adapted with SUnsET generate more relevant and factually consistent evidence with their summaries, extract evidence from more diverse locations in their context, and can generate more relevant and consistent summaries than baselines with no fine-tuning and fixed granularity evidence. We release SUnsET and our generation code to the public.

comment: EMNLP 2025 Main; 29 pages; 24 figures; 8 tables

♻ ☆ Epistemic Diversity and Knowledge Collapse in Large Language Models

Dustin Wright, Sarah Masud, Jared Moore, Srishti Yadav, Maria Antoniak, Chan Young Park, Isabelle Augenstein

Large language models (LLMs) tend to generate lexically, semantically, and stylistically homogenous texts. This poses a risk of knowledge collapse, where homogenous LLMs mediate a shrinking in the range of accessible information over time. Existing works on homogenization are limited by a focus on closed-ended multiple-choice setups or fuzzy semantic features, and do not look at trends across time and cultural contexts. To overcome this, we present a new methodology to measure epistemic diversity, i.e., variation in real-world claims in LLM outputs, which we use to perform a broad empirical study of LLM knowledge collapse. We test 27 LLMs, 155 topics covering 12 countries, and 200 prompt variations sourced from real user chats. For the topics in our study, we show that while newer models tend to generate more diverse claims, nearly all models are less epistemically diverse than a basic web search. We find that model size has a negative impact on epistemic diversity, while retrieval-augmented generation (RAG) has a positive impact, though the improvement from RAG varies by the cultural context. Finally, compared to a traditional knowledge source (Wikipedia), we find that country-specific claims reflect the English language more than the local one, highlighting a gap in epistemic representation

comment: 16 pages; 8 figures, 4 tables; v2 changelog: Fixed the modeling for table 3, random effect is the model version; v3 changelog: Fixed minor formatting issues in tables 2 and 3; v4 changelog: Fixed some typos and model description

♻ ☆ Dependency Structure Augmented Contextual Scoping Framework for Multimodal Aspect-Based Sentiment Analysis

Hao Liu, Lijun He, Jiaxi Liang, Zhihan Ren, Haixia Bi, Fan Li

Multimodal Aspect-Based Sentiment Analysis (MABSA) seeks to extract fine-grained information from image-text pairs to identify aspect terms and determine their sentiment polarity. However, existing approaches often fall short in simultaneously addressing three core challenges: Sentiment Cue Perception (SCP), Multimodal Information Misalignment (MIM), and Semantic Noise Elimination (SNE). To overcome these limitations, we propose DASCO (\textbf{D}ependency Structure \textbf{A}ugmented \textbf{Sco}ping Framework), a fine-grained scope-oriented framework that enhances aspect-level sentiment reasoning by leveraging dependency parsing trees. First, we designed a multi-task pretraining strategy for MABSA on our base model, combining aspect-oriented enhancement, image-text matching, and aspect-level sentiment-sensitive cognition. This improved the model's perception of aspect terms and sentiment cues while achieving effective image-text alignment, addressing key challenges like SCP and MIM. Furthermore, we incorporate dependency trees as syntactic branch combining with semantic branch, guiding the model to selectively attend to critical contextual elements within a target-specific scope while effectively filtering out irrelevant noise for addressing SNE problem. Extensive experiments on two benchmark datasets across three subtasks demonstrate that DASCO achieves state-of-the-art performance in MABSA, with notable gains in JMASA (+2.3\% F1 and +3.5\% precision on Twitter2015). The source code is available at https://github.com/LHaoooo/DASCO .

♻ ☆ PairUni: Pairwise Training for Unified Multimodal Language Models

Jiani Zheng, Zhiyang Teng, Xiangtai Li, Anran Wang, Yu Tian, Kunpeng Qiu, Ye Tian, Haochen Wang, Zhuochen Wang

Unified vision-language models (UVLMs) must perform both understanding and generation within a single architecture, but these tasks rely on heterogeneous data and supervision, making it difficult to balance them during reinforcement learning (RL). We propose PairUni, a unified framework that reorganizes data into understanding-generation (UG) pairs and aligns optimization accordingly. We first use GPT-o3 to augment single-task data, generating captions for understanding samples and question-answer (QA) pairs for generation samples, forming aligned pairs from the same instance. Additionally, for each generation sample, we retrieve a semantically related understanding example to form a retrieved pair, linking different but related data points. These paired structures expose cross-task semantic correspondences and support consistent policy learning. To leverage this structure, we present Pair-GPRO, a pair-aware variant based on Group Relative Policy Optimization. It assigns a similarity score to each pair to modulate the advantage, strengthening learning from well-aligned examples and reducing task interference. We curate a high-quality dataset of 16K UG pairs named PairUG for RL fine-tuning and evaluate PairUni on the powerful Janus-Pro UVLMs. Our approach achieves balanced improvements on various UVLMs, outperforming strong UVLM RL baselines. Codes are available at https://github.com/Haochen-Wang409/PairUni.

comment: 21 pages, 11 figures, and 8 tables

♻ ☆ Evaluating the Role of Verifiers in Test-Time Scaling for Legal Reasoning Tasks EMNLP

Davide Romano, Jonathan Schwarz, Daniele Giofré

Test-time scaling (TTS) techniques can improve the performance of large language models (LLMs) at the expense of additional computation and latency. While TTS has proven effective in formal domains such as mathematics and programming, its value in argumentative domains such as law remains underexplored. We present an empirical study of verifier-based TTS methods for legal multiple-choice QA (MCQA) across five benchmarks. Using a family of 7 reward models, we evaluate both outcome-level (Best-of-$N$) and process-level (tree search) verification under realistic low-$N$ budgets. Our analysis systematically investigates how verifier utility is affected by key properties such as domain specialization, model size, and supervision type (process-supervised PRMs vs. outcome-only ORMs), even when applied across different roles.

comment: Accepted to EMNLP - NLLP Workshop

♻ ☆ More of the Same: Persistent Representational Harms Under Increased Representation NeurIPS 2025

Jennifer Mickel, Maria De-Arteaga, Leqi Liu, Kevin Tian

To recognize and mitigate the harms of generative AI systems, it is crucial to consider who is represented in the outputs of generative AI systems and how people are represented. A critical gap emerges when naively improving who is represented, as this does not imply bias mitigation efforts have been applied to address how people are represented. We critically examined this by investigating gender representation in occupation across state-of-the-art large language models. We first show evidence suggesting that over time there have been interventions to models altering the resulting gender distribution, and we find that women are more represented than men when models are prompted to generate biographies or personas. We then demonstrate that representational biases persist in how different genders are represented by examining statistically significant word differences across genders. This results in a proliferation of representational harms, stereotypes, and neoliberalism ideals that, despite existing interventions to increase female representation, reinforce existing systems of oppression.

comment: Accepted by the 39th Conference on Neural Information Processing Systems (NeurIPS 2025) as a poster paper; 39 pages, 7 figures, 15 tables

♻ ☆ MedAgentBoard: Benchmarking Multi-Agent Collaboration with Conventional Methods for Diverse Medical Tasks NeurIPS 2025

Yinghao Zhu, Ziyi He, Haoran Hu, Xiaochen Zheng, Xichen Zhang, Zixiang Wang, Junyi Gao, Liantao Ma, Lequan Yu

The rapid advancement of Large Language Models (LLMs) has stimulated interest in multi-agent collaboration for addressing complex medical tasks. However, the practical advantages of multi-agent collaboration approaches remain insufficiently understood. Existing evaluations often lack generalizability, failing to cover diverse tasks reflective of real-world clinical practice, and frequently omit rigorous comparisons against both single-LLM-based and established conventional methods. To address this critical gap, we introduce MedAgentBoard, a comprehensive benchmark for the systematic evaluation of multi-agent collaboration, single-LLM, and conventional approaches. MedAgentBoard encompasses four diverse medical task categories: (1) medical (visual) question answering, (2) lay summary generation, (3) structured Electronic Health Record (EHR) predictive modeling, and (4) clinical workflow automation, across text, medical images, and structured EHR data. Our extensive experiments reveal a nuanced landscape: while multi-agent collaboration demonstrates benefits in specific scenarios, such as enhancing task completeness in clinical workflow automation, it does not consistently outperform advanced single LLMs (e.g., in textual medical QA) or, critically, specialized conventional methods that generally maintain better performance in tasks like medical VQA and EHR-based prediction. MedAgentBoard offers a vital resource and actionable insights, emphasizing the necessity of a task-specific, evidence-based approach to selecting and developing AI solutions in medicine. It underscores that the inherent complexity and overhead of multi-agent collaboration must be carefully weighed against tangible performance gains. All code, datasets, detailed prompts, and experimental results are open-sourced at https://medagentboard.netlify.app/.

comment: Accepted by NeurIPS 2025 Datasets & Benchmarks Track

♻ ☆ Wisdom and Delusion of LLM Ensembles for Code Generation and Repair

Fernando Vallecillos-Ruiz, Max Hort, Leon Moonen

Today's pursuit of a single Large Language Model (LMM) for all software engineering tasks is resource-intensive and overlooks the potential benefits of complementarity, where different models contribute unique strengths. However, the degree to which coding LLMs complement each other and the best strategy for maximizing an ensemble's potential are unclear, leaving practitioners without a clear path to move beyond single-model systems. To address this gap, we empirically compare ten individual LLMs from five families, and three ensembles of these LLMs across three software engineering benchmarks covering code generation and program repair. We assess the complementarity between models and the performance gap between the best individual model and the ensembles. Next, we evaluate various selection heuristics to identify correct solutions from an ensemble's candidate pool. We find that the theoretical upperbound for an ensemble's performance can be 83% above the best single model. Our results show that consensus-based strategies for selecting solutions fall into a "popularity trap," amplifying common but incorrect outputs. In contrast, a diversity-based strategy realizes up to 95% of this theoretical potential, and proves effective even in small two-model ensembles, enabling a cost-efficient way to enhance performance by leveraging multiple LLMs.

comment: Added Acknowledgments section and hyphenated last names

♻ ☆ TwinVoice: A Multi-dimensional Benchmark Towards Digital Twins via LLM Persona Simulation

Bangde Du, Minghao Guo, Songming He, Ziyi Ye, Xi Zhu, Weihang Su, Shuqi Zhu, Yujia Zhou, Yongfeng Zhang, Qingyao Ai, Yiqun Liu

Large Language Models (LLMs) are exhibiting emergent human-like abilities and are increasingly envisioned as the foundation for simulating an individual's communication style, behavioral tendencies, and personality traits. However, current evaluations of LLM-based persona simulation remain limited: most rely on synthetic dialogues, lack systematic frameworks, and lack analysis of the capability requirement. To address these limitations, we introduce TwinVoice, a comprehensive benchmark for assessing persona simulation across diverse real-world contexts. TwinVoice encompasses three dimensions: Social Persona (public social interactions), Interpersonal Persona (private dialogues), and Narrative Persona (role-based expression). It further decomposes the evaluation of LLM performance into six fundamental capabilities, including opinion consistency, memory recall, logical reasoning, lexical fidelity, persona tone, and syntactic style. Experimental results reveal that while advanced models achieve moderate accuracy in persona simulation, they still fall short of capabilities such as syntactic style and memory recall. Consequently, the average performance achieved by LLMs remains considerably below the human baseline.

comment: Main paper: 11 pages, 3 figures, 6 tables. Appendix: 28 pages. Bangde Du and Minghao Guo contributed equally. Corresponding authors: Ziyi Ye (ziyiye@fudan.edu.cn), Qingyao Ai (aiqy@tsinghua.edu.cn)

♻ ☆ ReForm: Reflective Autoformalization with Prospective Bounded Sequence Optimization

Guoxin Chen, Jing Wu, Xinjie Chen, Wayne Xin Zhao, Ruihua Song, Chengxi Li, Kai Fan, Dayiheng Liu, Minpeng Liao

Autoformalization, which translates natural language mathematics into machine-verifiable formal statements, is critical for using formal mathematical reasoning to solve math problems stated in natural language. While Large Language Models can generate syntactically correct formal statements, they often fail to preserve the original problem's semantic intent. This limitation arises from the LLM approaches' treating autoformalization as a simplistic translation task which lacks mechanisms for self-reflection and iterative refinement that human experts naturally employ. To address these issues, we propose ReForm, a Reflective Autoformalization method that tightly integrates semantic consistency evaluation into the autoformalization process. This enables the model to iteratively generate formal statements, assess its semantic fidelity, and self-correct identified errors through progressive refinement. To effectively train this reflective model, we introduce Prospective Bounded Sequence Optimization (PBSO), which employs different rewards at different sequence positions to ensure that the model develops both accurate autoformalization and correct semantic validations, preventing superficial critiques that would undermine the purpose of reflection. Extensive experiments across four autoformalization benchmarks demonstrate that ReForm achieves an average improvement of 22.6 percentage points over the strongest baselines. To further ensure evaluation reliability, we introduce ConsistencyCheck, a benchmark of 859 expert-annotated items that not only validates LLMs as judges but also reveals that autoformalization is inherently difficult: even human experts produce semantic errors in up to 38.5% of cases.

comment: https://github.com/Chen-GX/ReForm

♻ ☆ Towards a Method for Synthetic Generation of Persons with Aphasia Transcripts

Jason M. Pittman, Anton Phillips Jr., Yesenia Medina-Santos, Brielle C. Stark

In aphasia research, Speech-Language Pathologists (SLPs) devote extensive time to manually coding speech samples using Correct Information Units (CIUs), a measure of how informative an individual sample of speech is. Developing automated systems to recognize aphasic language is limited by data scarcity. For example, only about 600 transcripts are available in AphasiaBank yet billions of tokens are used to train large language models (LLMs). In the broader field of machine learning (ML), researchers increasingly turn to synthetic data when such are sparse. Therefore, this study constructs and validates two methods to generate synthetic transcripts of the AphasiaBank Cat Rescue picture description task. One method leverages a procedural programming approach while the second uses Mistral 7b Instruct and Llama 3.1 8b Instruct LLMs. The methods generate transcripts across four severity levels (Mild, Moderate, Severe, Very Severe) through word dropping, filler insertion, and paraphasia substitution. Overall, we found, compared to human-elicited transcripts, Mistral 7b Instruct best captures key aspects of linguistic degradation observed in aphasia, showing realistic directional changes in NDW, word count, and word length amongst the synthetic generation methods. Based on the results, future work should plan to create a larger dataset, fine-tune models for better aphasic representation, and have SLPs assess the realism and usefulness of the synthetic transcripts.

comment: 19 pages, 1 figure, 7 tables

♻ ☆ Paper2Poster: Towards Multimodal Poster Automation from Scientific Papers

Wei Pang, Kevin Qinghong Lin, Xiangru Jian, Xi He, Philip Torr

Academic poster generation is a crucial yet challenging task in scientific communication, requiring the compression of long-context interleaved documents into a single, visually coherent page. To address this challenge, we introduce the first benchmark and metric suite for poster generation, which pairs recent conference papers with author-designed posters and evaluates outputs on (i)Visual Quality-semantic alignment with human posters, (ii)Textual Coherence-language fluency, (iii)Holistic Assessment-six fine-grained aesthetic and informational criteria scored by a VLM-as-judge, and notably (iv)PaperQuiz-the poster's ability to convey core paper content as measured by VLMs answering generated quizzes. Building on this benchmark, we propose PosterAgent, a top-down, visual-in-the-loop multi-agent pipeline: the (a)Parser distills the paper into a structured asset library; the (b)Planner aligns text-visual pairs into a binary-tree layout that preserves reading order and spatial balance; and the (c)Painter-Commenter loop refines each panel by executing rendering code and using VLM feedback to eliminate overflow and ensure alignment. In our comprehensive evaluation, we find that GPT-4o outputs-though visually appealing at first glance-often exhibit noisy text and poor PaperQuiz scores, and we find that reader engagement is the primary aesthetic bottleneck, as human-designed posters rely largely on visual semantics to convey meaning. Our fully open-source variants (e.g. based on the Qwen-2.5 series) outperform existing 4o-driven multi-agent systems across nearly all metrics, while using 87% fewer tokens. It transforms a 22-page paper into a finalized yet editable .pptx poster - all for just $0.005. These findings chart clear directions for the next generation of fully automated poster-generation models. The code and datasets are available at https://github.com/Paper2Poster/Paper2Poster.

comment: Project Page: https://github.com/Paper2Poster/Paper2Poster

♻ ☆ BhashaBench V1: A Comprehensive Benchmark for the Quadrant of Indic Domains

Vijay Devane, Mohd Nauman, Bhargav Patel, Aniket Mahendra Wakchoure, Yogeshkumar Sant, Shyam Pawar, Viraj Thakur, Ananya Godse, Sunil Patra, Neha Maurya, Suraj Racha, Nitish Kamal Singh, Ajay Nagpal, Piyush Sawarkar, Kundeshwar Vijayrao Pundalik, Rohit Saluja, Ganesh Ramakrishnan

The rapid advancement of large language models(LLMs) has intensified the need for domain and culture specific evaluation. Existing benchmarks are largely Anglocentric and domain-agnostic, limiting their applicability to India-centric contexts. To address this gap, we introduce BhashaBench V1, the first domain-specific, multi-task, bilingual benchmark focusing on critical Indic knowledge systems. BhashaBench V1 contains 74,166 meticulously curated question-answer pairs, with 52,494 in English and 21,672 in Hindi, sourced from authentic government and domain-specific exams. It spans four major domains: Agriculture, Legal, Finance, and Ayurveda, comprising 90+ subdomains and covering 500+ topics, enabling fine-grained evaluation. Evaluation of 29+ LLMs reveals significant domain and language specific performance gaps, with especially large disparities in low-resource domains. For instance, GPT-4o achieves 76.49% overall accuracy in Legal but only 59.74% in Ayurveda. Models consistently perform better on English content compared to Hindi across all domains. Subdomain-level analysis shows that areas such as Cyber Law, International Finance perform relatively well, while Panchakarma, Seed Science, and Human Rights remain notably weak. BhashaBench V1 provides a comprehensive dataset for evaluating large language models across India's diverse knowledge domains. It enables assessment of models' ability to integrate domain-specific knowledge with bilingual understanding. All code, benchmarks, and resources are publicly available to support open research.

♻ ☆ MindGYM: What Matters in Question Synthesis for Thinking-Centric Fine-Tuning? NeurIPS'25

Zhe Xu, Daoyuan Chen, Zhenqing Ling, Yaliang Li, Ying Shen

Large foundation models face challenges in acquiring transferable, structured thinking abilities, especially when supervised with rigid templates or crowd-annotated instruction datasets. Unlike prior approaches, we focus on a thinking-centric data synthesis paradigm that enables models to evolve through self-generated, cognitively guided data. We propose MindGYM, a structured and scalable framework for question synthesis, composed of: (1) Cognitive Thinking Process Injection, which infuses high-level reasoning objectives to shape the model's synthesis behavior; (2) Seed Single-Hop Question Synthesis, generating atomic questions from diverse semantic types to encourage broader thinking; and (3) Challenging Multi-Hop QA Synthesis, composing more complex multi-hop questions based on QA seeds for deeper reasoning. Detailed analysis shows that synthetic data generated by our method achieves 16.7% higher average quality and 67.91% lower quality variance compared to baseline sources, highlighting that both high-quality and self-contained data are essential for effective, thinking-oriented fine-tuning. MindGYM improves performance on six reasoning benchmarks, achieving gains of up to 16% on MathVision using only 400 data samples, and generalizable improvements across different model sizes and architectures. MindGYM underscores the viability of self-challenging mechanisms in refining large model capabilities while minimizing human intervention and resource demands. Code and data are released to promote data-centric research into self-evolving foundation models driven by their internal reasoning capabilities.

comment: Accepted by NeurIPS'25. 30 pages, 2 figures, 13 tables

♻ ☆ UNO-Bench: A Unified Benchmark for Exploring the Compositional Law Between Uni-modal and Omni-modal in Omni Models

Chen Chen, ZeYang Hu, Fengjiao Chen, Liya Ma, Jiaxing Liu, Xiaoyu Li, Ziwen Wang, Xuezhi Cao, Xunliang Cai

Multimodal Large Languages models have been progressing from uni-modal understanding toward unifying visual, audio and language modalities, collectively termed omni models. However, the correlation between uni-modal and omni-modal remains unclear, which requires comprehensive evaluation to drive omni model's intelligence evolution. In this work, we introduce a novel, high-quality, and UNified Omni model benchmark, UNO-Bench. This benchmark is designed to effectively evaluate both UNi-modal and Omni-modal capabilities under a unified ability taxonomy, spanning 44 task types and 5 modality combinations. It includes 1250 human curated samples for omni-modal with 98% cross-modality solvability, and 2480 enhanced uni-modal samples. The human-generated dataset is well-suited to real-world scenarios, particularly within the Chinese context, whereas the automatically compressed dataset offers a 90% increase in speed and maintains 98% consistency across 18 public benchmarks. In addition to traditional multi-choice questions, we propose an innovative multi-step open-ended question format to assess complex reasoning. A general scoring model is incorporated, supporting 6 question types for automated evaluation with 95% accuracy. Experimental result shows the Compositional Law between omni-modal and uni-modal performance and the omni-modal capability manifests as a bottleneck effect on weak models, while exhibiting synergistic promotion on strong models.

comment: v3: Switch the paper template. Work in progress. Github: https://github.com/meituan-longcat/UNO-Bench Hugging Face: https://huggingface.co/datasets/meituan-longcat/UNO-Bench

♻ ☆ Beyond Isolated Dots: Benchmarking Structured Table Construction as Deep Knowledge Extraction

Tianyun Zhong, Guozhao Mo, Yanjiang Liu, Yihan Chen, Lingdi Kong, Xuanang Chen, Yaojie Lu, Hongyu Lin, Shiwei Ye, Xianpei Han, Ben He, Le Sun

With the emergence of large language models (LLMs), there is an expectation that LLMs can effectively extract explicit information from complex real-world documents (e.g., papers, reports). However, most LLMs generate paragraph-style answers that are chaotic, disorganized, and untraceable. To bridge this gap, we introduce the Arranged and Organized Extraction Benchmark (AOE), a new bilingual benchmark with data and documents of varying lengths designed to systematically evaluate the ability of LLMs to comprehend fragmented documents and reconstruct isolated information into one organized table. Unlike conventional text-to-table tasks, which rely on fixed schema and narrow task domains, AOE includes 11 carefully crafted tasks across three diverse domains, requiring models to generate context-specific schema tailored to varied input queries. In the experiment, we evaluated both open-source and closed-source state-of-the-art LLMs. The results show that even the most advanced models struggled significantly. The benchmark is available at https://anonymous.4open.science/r/AOE-Benchmark/.

♻ ☆ Large Language Models Have Intrinsic Meta-Cognition, but Need a Good Lens EMNLP 2025

Ziyang Ma, Qingyue Yuan, Zhenglin Wang, Deyu Zhou

Previous research has primarily focused on the cognitive error detection capabilities of Large Language Models (LLMs), often prompting them to analyze mistakes in reasoning chains. However, few studies have examined the meta-cognitive abilities of LLMs (e.g., their self-awareness of step errors), which are crucial for their reliability. While studies on LLM self-evaluation present some measures, such as perplexity, which can reflect the answer correctness and be viewed as the lens of meta-cognition, they lack step-level analysis and adaptation. This paper studies the evaluation of LLM meta-cognition using the current lenses and how to improve these lenses. Specifically, we propose AutoMeco, an Automated Meta-cognition Evaluation framework for benchmarking the existing lenses. Furthermore, a training-free Markovian Intrinsic Reward Adjustment strategy, MIRA, is proposed to boost current meta-cognition lenses. Experimental results on three mathematical reasoning datasets and three LLMs show the reasonableness of AutoMeco by comparing it with Best-of-N verification. Moreover, the meta-cognition ability of LLMs can be better evaluated using MIRA.

comment: Accepted to EMNLP 2025

♻ ☆ Hysteresis Activation Function for Efficient Inference NeurIPS

Moshe Kimhi, Idan Kashani, Avi Mendelson, Chaim Baskin

The widely used ReLU is favored for its hardware efficiency, {as the implementation at inference is a one bit sign case,} yet suffers from issues such as the ``dying ReLU'' problem, where during training, neurons fail to activate and constantly remain at zero, as highlighted by Lu et al. Traditional approaches to mitigate this issue often introduce more complex and less hardware-friendly activation functions. In this work, we propose a Hysteresis Rectified Linear Unit (HeLU), an efficient activation function designed to address the ``dying ReLU'' problem with minimal complexity. Unlike traditional activation functions with fixed thresholds for training and inference, HeLU employs a variable threshold that refines the backpropagation. This refined mechanism allows simpler activation functions to achieve competitive performance comparable to their more complex counterparts without introducing unnecessary complexity or requiring inductive biases. Empirical evaluations demonstrate that HeLU enhances model generalization across diverse datasets, offering a promising solution for efficient and effective inference suitable for a wide range of neural network architectures.

comment: Accepted to 4th NeurIPS Efficient Natural Language and Speech Processing Workshop (ENLSP-IV 2024)

♻ ☆ Diversity as a Reward: Fine-Tuning LLMs on a Mixture of Domain-Undetermined Data NeurIPS'25

Zhenqing Ling, Daoyuan Chen, Liuyi Yao, Qianli Shen, Yaliang Li, Ying Shen

Fine-tuning large language models (LLMs) using diverse datasets is crucial for enhancing their overall performance across various domains. In practical scenarios, existing methods based on modeling the mixture proportions of data composition often struggle with data whose domain labels are missing, imprecise or non-normalized, while methods based on data selection usually encounter difficulties in balancing multi-domain performance. To address these challenges, in this work, we investigate the role of data diversity in enhancing the overall abilities of LLMs by empirically constructing contrastive data pools and theoretically deriving explanations. Building upon the insights gained, we propose a new method that gives the LLM a dual identity: an output model to cognitively probe and select data based on diversity reward, as well as an input model to be tuned with the selected data. Extensive experiments show that the proposed method notably boosts performance across domain-undetermined data and a series of foundational downstream tasks when applied to various advanced LLMs. We release our code and hope this study can shed light on the understanding of data diversity and advance feedback-driven data-model co-design for LLMs.

comment: Accepted by NeurIPS'25 main track. 47 pages, 21 figures, 32 tables

♻ ☆ SEA-LION: Southeast Asian Languages in One Network AACL 2025

Raymond Ng, Thanh Ngan Nguyen, Yuli Huang, Ngee Chia Tai, Wai Yi Leong, Wei Qi Leong, Xianbin Yong, Jian Gang Ngui, Yosephine Susanto, Nicholas Cheng, Hamsawardhini Rengarajan, Peerat Limkonchotiwat, Adithya Venkatadri Hulagadri, Kok Wai Teng, Yeo Yeow Tong, Bryan Siow, Wei Yi Teo, Wayne Lau, Choon Meng Tan, Brandon Ong, Zhi Hao Ong, Jann Railey Montalan, Adwin Chan, Sajeban Antonyrex, Ren Lee, Esther Choa, David Ong Tat-Wee, Bing Jie Darius Liu, William Chandra Tjhi, Erik Cambria, Leslie Teo

Recently, Large Language Models (LLMs) have dominated much of the artificial intelligence scene with their ability to process and generate natural languages. However, the majority of LLM research and development remains English-centric, leaving low-resource languages such as those in the Southeast Asian (SEA) region under-represented. To address this representation gap, we introduce Llama-SEA-LION-v3-8B-IT and Gemma-SEA-LION-v3-9B-IT, two cutting-edge multilingual LLMs designed for SEA languages. The SEA-LION family of LLMs supports 11 SEA languages, namely English, Chinese, Indonesian, Vietnamese, Malay, Thai, Burmese, Lao, Filipino, Tamil, and Khmer. Our work leverages large-scale multilingual continued pre-training with a comprehensive post-training regime involving multiple stages of instruction fine-tuning, alignment, and model merging. Evaluation results on multilingual benchmarks indicate that our models achieve state-of-the-art performance across LLMs supporting SEA languages. We open-source the models to benefit the wider SEA community.

comment: Accepted at IJCNLP-AACL 2025 (Main Track). We released our model at https://huggingface.co/collections/aisingapore/sea-lionv3-672589a39cdadd6a5b199581

♻ ☆ Model-Document Protocol for AI Search

Hongjin Qian, Zheng Liu

AI search depends on linking large language models (LLMs) with vast external knowledge sources. Yet web pages, PDF files, and other raw documents are not inherently LLM-ready: they are long, noisy, and unstructured. Conventional retrieval methods treat these documents as verbatim text and return raw passages, leaving the burden of fragment assembly and contextual reasoning to the LLM. This gap underscores the need for a new retrieval paradigm that redefines how models interact with documents. We introduce the Model-Document Protocol (MDP), a general framework that formalizes how raw text is bridged to LLMs through consumable knowledge representations. Rather than treating retrieval as passage fetching, MDP defines multiple pathways that transform unstructured documents into task-specific, LLM-ready inputs. These include agentic reasoning, which curates raw evidence into coherent context; memory grounding, which accumulates reusable notes to enrich reasoning; and structured leveraging, which encodes documents into formal representations such as graphs or key-value caches. All three pathways share the same goal: ensuring that what reaches the LLM is not raw fragments but compact, structured knowledge directly consumable for reasoning. As an instantiation, we present MDP-Agent, which realizes the protocol through an agentic process: constructing document-level gist memories for global coverage, performing diffusion-based exploration with vertical exploitation to uncover layered dependencies, and applying map-reduce style synthesis to integrate large-scale evidence into compact yet sufficient context. Experiments on information-seeking benchmarks demonstrate that MDP-Agent outperforms baselines, validating both the soundness of the MDP framework and the effectiveness of its agentic instantiation.

comment: 10 pages

♻ ☆ How Efficient Are Diffusion Language Models? A Critical Examination of Efficiency Evaluation Practices

Han Peng, Peiyu Liu, Zican Dong, Daixuan Cheng, Junyi Li, Yiru Tang, Shuo Wang, Wayne Xin Zhao

Diffusion language models (DLMs) have emerged as a promising alternative to the long-dominant autoregressive (AR) paradigm, offering a parallelable decoding process that could yield greater efficiency. Yet, in practice, current open-source DLMs often underperform their AR counterparts in speed, limiting their real-world utility. This work presents a systematic study of DLM efficiency, identifying key issues in prior evaluation methods. Through empirical benchmarking and a roofline-based theoretical analysis, we demonstrate that AR models generally achieve higher throughput, while DLMs consistently lag. We also investigate acceleration strategies, finding that techniques like dual cache and parallel decoding mainly offer gains at small batch sizes, with their benefits diminishing upon scaling. Our findings underscore the necessity of robust evaluation methods and improved acceleration strategies to advance research on DLMs.

comment: Withdrawn by the authors to better delineate the related work from the paper's original contributions

♻ ☆ SPARTA ALIGNMENT: Collectively Aligning Multiple Language Models through Combat NeurIPS 2025

Yuru Jiang, Wenxuan Ding, Shangbin Feng, Greg Durrett, Yulia Tsvetkov

We propose SPARTA ALIGNMENT, an algorithm to collectively align multiple LLMs through competition and combat. To complement a single model's lack of diversity in generation and biases in evaluation, multiple LLMs form a "sparta tribe" to compete against each other in fulfilling instructions while serving as judges for the competition of others. For each iteration, one instruction and two models are selected for a duel, the other models evaluate the two responses, and their evaluation scores are aggregated through a adapted elo-ranking based reputation system, where winners/losers of combat gain/lose weight in evaluating others. The peer-evaluated combat results then become preference pairs where the winning response is preferred over the losing one, and all models learn from these preferences at the end of each iteration. SPARTA ALIGNMENT enables the self-evolution of multiple LLMs in an iterative and collective competition process. Extensive experiments demonstrate that SPARTA ALIGNMENT outperforms initial models and 4 self-alignment baselines across 10 out of 12 tasks and datasets with 7.0% average improvement. Further analysis reveals that SPARTA ALIGNMENT generalizes more effectively to unseen tasks and leverages the expertise diversity of participating models to produce more logical, direct and informative outputs.

comment: NeurIPS 2025

♻ ☆ IGD: Token Decisiveness Modeling via Information Gain in LLMs for Personalized Recommendation

Zijie Lin, Yang Zhang, Xiaoyan Zhao, Fengbin Zhu, Fuli Feng, Tat-Seng Chua

Large Language Models (LLMs) have shown strong potential for recommendation by framing item prediction as a token-by-token language generation task. However, existing methods treat all item tokens equally, simply pursuing likelihood maximization during both optimization and decoding. This overlooks crucial token-level differences in decisiveness-many tokens contribute little to item discrimination yet can dominate optimization or decoding. To quantify token decisiveness, we propose a novel perspective that models item generation as a decision process, measuring token decisiveness by the Information Gain (IG) each token provides in reducing uncertainty about the generated item. Our empirical analysis reveals that most tokens have low IG but often correspond to high logits, disproportionately influencing training loss and decoding, which may impair model performance. Building on these insights, we introduce an Information Gain-based Decisiveness-aware Token handling (IGD) strategy that integrates token decisiveness into both tuning and decoding. Specifically, IGD downweights low-IG tokens during tuning and rebalances decoding to emphasize tokens with high IG. In this way, IGD moves beyond pure likelihood maximization, effectively prioritizing high-decisiveness tokens. Extensive experiments on four benchmark datasets with two LLM backbones demonstrate that IGD consistently improves recommendation accuracy, achieving significant gains on widely used ranking metrics compared to strong baselines.

♻ ☆ ClueAnchor: Clue-Anchored Knowledge Reasoning Exploration and Optimization for Retrieval-Augmented Generation

Hao Chen, Yukun Yan, Sen Mei, Wanxiang Che, Zhenghao Liu, Qi Shi, Xinze Li, Yuchun Fan, Pengcheng Huang, Qiushi Xiong, Zhiyuan Liu, Maosong Sun

Retrieval-Augmented Generation (RAG) augments Large Language Models (LLMs) with external knowledge to improve factuality. However, existing RAG systems frequently underutilize the retrieved documents, failing to extract and integrate the key clues needed to support faithful and interpretable reasoning, especially in cases where relevant evidence is implicit, scattered, or obscured by noise. To address this issue, we propose ClueAnchor, a novel framework for enhancing RAG via clue-anchored reasoning exploration and optimization. ClueAnchor extracts key clues from retrieved content and generates multiple reasoning paths based on different knowledge configurations, optimizing the model by selecting the most appropriate reasoning path for the given context through reward-based preference optimization. Experiments show that ClueAnchor significantly outperforms prior RAG baselines in the completeness and robustness of reasoning. Further analysis confirms its strong resilience to noisy or partially relevant retrieved content, as well as its capability to identify supporting evidence even in the absence of explicit clue supervision during inference. All codes are available at https://github.com/thunlp/ClueAnchor.

♻ ☆ Nek Minit: Harnessing Pragmatic Metacognitive Prompting for Explainable Sarcasm Detection of Australian and Indian English ALT

Ishmanbir Singh, Dipankar Srirag, Aditya Joshi

Sarcasm is a challenge to sentiment analysis because of the incongruity between stated and implied sentiment. The challenge is exacerbated when the implication may be relevant to a specific country or geographical region. Pragmatic metacognitive prompting (PMP) is a cognition-inspired technique that has been used for pragmatic reasoning. In this paper, we harness PMP for explainable sarcasm detection for Australian and Indian English, alongside a benchmark dataset for standard English. We manually add sarcasm explanations to an existing sarcasm-labeled dataset for Australian and Indian English called BESSTIE, and compare the performance for explainable sarcasm detection for them with FLUTE, a standard English dataset containing sarcasm explanations. Our approach utilising PMP when evaluated on two open-weight LLMs (GEMMA and LLAMA) achieves statistically significant performance improvement across all tasks and datasets when compared with four alternative prompting strategies. We also find that alternative techniques such as agentic prompting mitigate context-related failures by enabling external knowledge retrieval. The focused contribution of our work is utilising PMP in generating sarcasm explanations for varieties of English.

comment: ALTA 2025 (Best Paper Honorable Mention). Camera-ready

♻ ☆ FESTA: Functionally Equivalent Sampling for Trust Assessment of Multimodal LLMs EMNLP

Debarpan Bhattacharya, Apoorva Kulkarni, Sriram Ganapathy

The accurate trust assessment of multimodal large language models (MLLMs) generated predictions, which can enable selective prediction and improve user confidence, is challenging due to the diverse multi-modal input paradigms. We propose Functionally Equivalent Sampling for Trust Assessment (FESTA), a multimodal input sampling technique for MLLMs, that generates an uncertainty measure based on the equivalent and complementary input samplings. The proposed task-preserving sampling approach for uncertainty quantification expands the input space to probe the consistency (through equivalent samples) and sensitivity (through complementary samples) of the model. FESTA uses only input-output access of the model (black-box), and does not require ground truth (unsupervised). The experiments are conducted with various off-the-shelf multi-modal LLMs, on both visual and audio reasoning tasks. The proposed FESTA uncertainty estimate achieves significant improvement (33.3% relative improvement for vision-LLMs and 29.6% relative improvement for audio-LLMs) in selective prediction performance, based on area-under-receiver-operating-characteristic curve (AUROC) metric in detecting mispredictions. The code implementation is open-sourced.

comment: Accepted in the Findings of EMNLP, 2025

♻ ☆ Seek in the Dark: Reasoning via Test-Time Instance-Level Policy Gradient in Latent Space

Hengli Li, Chenxi Li, Tong Wu, Xuekai Zhu, Yuxuan Wang, Zhaoxin Yu, Eric Hanchen Jiang, Song-Chun Zhu, Zixia Jia, Ying Nian Wu, Zilong Zheng

Reasoning ability, a core component of human intelligence, continues to pose a significant challenge for Large Language Models (LLMs) in the pursuit of AGI. Although model performance has improved under the training scaling law, significant challenges remain, particularly with respect to training algorithms, such as catastrophic forgetting, and the limited availability of novel training data. As an alternative, test-time scaling enhances reasoning performance by increasing test-time computation without parameter updating. Unlike prior methods in this paradigm focused on token space, we propose leveraging latent space for more effective reasoning and better adherence to the test-time scaling law. We introduce LatentSeek, a novel framework that enhances LLM reasoning through Test-Time Instance-level Adaptation (TTIA) within the model's latent space. Specifically, LatentSeek leverages policy gradient to iteratively update latent representations, guided by self-generated reward signals. LatentSeek is evaluated on a range of reasoning benchmarks, including GSM8K, MATH-500, and AIME2024, across multiple LLM architectures. Results show that LatentSeek consistently outperforms strong baselines, such as Chain-of-Thought prompting and fine-tuning-based methods. Furthermore, our analysis demonstrates that LatentSeek is highly efficient, typically converging within a few iterations for problems of average complexity, while also benefiting from additional iterations, thereby highlighting the potential of test-time scaling in the latent space. These findings position LatentSeek as a lightweight, scalable, and effective solution for enhancing the reasoning capabilities of LLMs.

♻ ☆ The Scales of Justitia: A Comprehensive Survey on Safety Evaluation of LLMs

Songyang Liu, Chaozhuo Li, Jiameng Qiu, Xi Zhang, Feiran Huang, Litian Zhang, Yiming Hei, Philip S. Yu

With the rapid advancement of artificial intelligence, Large Language Models (LLMs) have shown remarkable capabilities in Natural Language Processing (NLP), including content generation, human-computer interaction, machine translation, and code generation. However, their widespread deployment has also raised significant safety concerns. In particular, LLM-generated content can exhibit unsafe behaviors such as toxicity, bias, or misinformation, especially in adversarial contexts, which has attracted increasing attention from both academia and industry. Although numerous studies have attempted to evaluate these risks, a comprehensive and systematic survey on safety evaluation of LLMs is still lacking. This work aims to fill this gap by presenting a structured overview of recent advances in safety evaluation of LLMs. Specifically, we propose a four-dimensional taxonomy: (i) Why to evaluate, which explores the background of safety evaluation of LLMs, how they differ from general LLMs evaluation, and the significance of such evaluation; (ii) What to evaluate, which examines and categorizes existing safety evaluation tasks based on key capabilities, including dimensions such as toxicity, robustness, ethics, bias and fairness, truthfulness, and related aspects; (iii) Where to evaluate, which summarizes the evaluation metrics, datasets and benchmarks currently used in safety evaluations; (iv) How to evaluate, which reviews existing mainstream evaluation methods based on the roles of the evaluators and some evaluation frameworks that integrate the entire evaluation pipeline. Finally, we identify the challenges in safety evaluation of LLMs and propose promising research directions to promote further advancement in this field. We emphasize the necessity of prioritizing safety evaluation to ensure the reliable and responsible deployment of LLMs in real-world applications.

comment: 20 pages, preprint

♻ ☆ Similarity-Distance-Magnitude Activations

Allen Schmaltz

We introduce the Similarity-Distance-Magnitude (SDM) activation function, a more robust and interpretable formulation of the standard softmax activation function, adding Similarity (i.e., correctly predicted depth-matches into training) awareness and Distance-to-training-distribution awareness to the existing output Magnitude (i.e., decision-boundary) awareness, and enabling interpretability-by-exemplar via dense matching. We further introduce the SDM estimator, based on a data-driven partitioning of the class-wise empirical CDFs via the SDM activation, to control the class- and prediction-conditional accuracy among selective classifications. When used as the final-layer activation over pre-trained language models for selective classification, the SDM estimator is more robust to co-variate shifts and out-of-distribution inputs than existing calibration methods using softmax activations, while remaining informative over in-distribution data.

comment: 18 pages, 5 tables, 1 algorithm. arXiv admin note: substantial text overlap with arXiv:2502.20167

♻ ☆ TEXT2DB: Integration-Aware Information Extraction with Large Language Model Agents

Yizhu Jiao, Sha Li, Sizhe Zhou, Heng Ji, Jiawei Han

The task of information extraction (IE) is to extract structured knowledge from text. However, it is often not straightforward to utilize IE output due to the mismatch between the IE ontology and the downstream application needs. We propose a new formulation of IE TEXT2DB that emphasizes the integration of IE output and the target database (or knowledge base). Given a user instruction, a document set, and a database, our task requires the model to update the database with values from the document set to satisfy the user instruction. This task requires understanding user instructions for what to extract and adapting to the given DB/KB schema for how to extract on the fly. To evaluate this new task, we introduce a new benchmark featuring common demands such as data infilling, row population, and column addition. In addition, we propose an LLM agent framework OPAL (Observe-PlanAnalyze LLM) which includes an Observer component that interacts with the database, the Planner component that generates a code-based plan with calls to IE models, and the Analyzer component that provides feedback regarding code quality before execution. Experiments show that OPAL can successfully adapt to diverse database schemas by generating different code plans and calling the required IE models. We also highlight difficult cases such as dealing with large databases with complex dependencies and extraction hallucination, which we believe deserve further investigation. Source code: https://github.com/yzjiao/Text2DB

comment: Source code: https://github.com/yzjiao/Text2DB

♻ ☆ Towards Predicting Any Human Trajectory In Context NeurIPS 2025

Ryo Fujii, Hideo Saito, Ryo Hachiuma

Predicting accurate future trajectories of pedestrians is essential for autonomous systems but remains a challenging task due to the need for adaptability in different environments and domains. A common approach involves collecting scenario-specific data and performing fine-tuning via backpropagation. However, the need to fine-tune for each new scenario is often impractical for deployment on edge devices. To address this challenge, we introduce \paper, an In-Context Learning (ICL) framework for pedestrian trajectory prediction that enables adaptation without fine-tuning on the scenario-specific data at inference time without requiring weight updates. We propose a spatio-temporal similarity-based example selection (STES) method that selects relevant examples from previously observed trajectories within the same scene by identifying similar motion patterns at corresponding locations. To further refine this selection, we introduce prediction-guided example selection (PG-ES), which selects examples based on both the past trajectory and the predicted future trajectory, rather than relying solely on the past trajectory. This approach allows the model to account for long-term dynamics when selecting examples. Finally, instead of relying on small real-world datasets with limited scenario diversity, we train our model on a large-scale synthetic dataset to enhance its prediction ability by leveraging in-context examples. Extensive experiments demonstrate that TrajICL achieves remarkable adaptation across both in-domain and cross-domain scenarios, outperforming even fine-tuned approaches across multiple public benchmarks. Project Page: https://fujiry0.github.io/TrajICL-project-page/.

comment: NeurIPS 2025

♻ ☆ Speak & Spell: LLM-Driven Controllable Phonetic Error Augmentation for Robust Dialogue State Tracking AACL

Jihyun Lee, Solee Im, Wonjun Lee, Gary Geunbae Lee

Dialogue State Tracking (DST) is a key part of task-oriented dialogue systems, identifying important information in conversations. However, its accuracy drops significantly in spoken dialogue environments due to named entity errors from Automatic Speech Recognition (ASR) systems. We introduce a simple yet effective data augmentation method that targets those entities to improve the robustness of DST model. Our novel method can control the placement of errors using keyword-highlighted prompts while introducing phonetically similar errors. As a result, our method generated sufficient error patterns on keywords, leading to improved accuracy in noised and low-accuracy ASR environments.

comment: Accepted to AACL-IJCNLP 2025

♻ ☆ Are LLMs Rigorous Logical Reasoners? Empowering Natural Language Proof Generation by Stepwise Decoding with Contrastive Learning AACL 2025

Ying Su, Mingwen Liu, Zhijiang Guo

Logical reasoning is a pivotal component in the field of artificial intelligence. Proof planning, particularly in contexts requiring the validation of explanation accuracy, continues to present challenges. The recent advancement of large language models (LLMs) has led to significant progress in natural language proof planning, evolving from one-stage generators to more complex three-stage systems that include additional searchers or verifiers. While these assisted methods improve the quality of generated results, they also introduce increased search efforts and computational costs. Furthermore, the generative process itself remains underexplored. In this study, we propose a stepwise decoding approach augmented by contrastive learning to address two common errors encountered during the LLM generator's decoding process. We fine-tune the language model using both vanilla and enhanced hard negatives to mitigate these decoding errors. Empirical results demonstrate the effectiveness of our strategy. Additionally, our further analysis reveals that even larger LLMs still struggle to generate rigorous logical chains.

comment: 15 pages, 2 figures, 11 tables. Accepted by AACL 2025 main conference

♻ ☆ Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems

Christian Walder, Deep Karkhanis

Reinforcement Learning (RL) algorithms sample multiple n>1 solution attempts for each problem and reward them independently. This optimizes for pass@1 performance and prioritizes the strength of isolated samples at the expense of the diversity and collective utility of sets of samples. This under-utilizes the sampling capacity, limiting exploration and eventual improvement on harder examples. As a fix, we propose Pass-at-k Policy Optimization (PKPO), a transformation on the final rewards which leads to direct optimization of pass@k performance, thus optimizing for sets of samples that maximize reward when considered jointly. Our contribution is to derive novel low variance unbiased estimators for pass@k and its gradient, in both the binary and continuous reward settings. We show optimization with our estimators reduces to standard RL with rewards that have been jointly transformed by a stable and efficient transformation function. While previous efforts are restricted to k=n, ours is the first to enable robust optimization of pass@k for any arbitrary k <= n. Moreover, instead of trading off pass@1 performance for pass@k gains, our method allows annealing k during training, optimizing both metrics and often achieving strong pass@1 numbers alongside significant pass@k gains. We validate our reward transformations on toy experiments, which reveal the variance reducing properties of our formulations. We also include real-world examples using the open-source LLM, GEMMA-2. We find that our transformation effectively optimizes for the target k. Furthermore, higher k values enable solving more and harder problems, while annealing k boosts both the pass@1 and pass@k . Crucially, for challenging task sets where conventional pass@1 optimization stalls, our pass@k approach unblocks learning, likely due to better exploration by prioritizing joint utility over the utility of individual samples.

♻ ☆ Neural Networks for Learnable and Scalable Influence Estimation of Instruction Fine-Tuning Data

Ishika Agarwal, Dilek Hakkani-Tür

Influence functions provide crucial insights into model training, but existing methods suffer from large computational costs and limited generalization. Particularly, recent works have proposed various metrics and algorithms to calculate the influence of data using language models, which do not scale well with large models and datasets. This is because of the expensive forward and backward passes required for computation, substantial memory requirements to store large models, and poor generalization of influence estimates to new data. In this paper, we explore the use of small neural networks -- which we refer to as the InfluenceNetwork -- to estimate influence values, achieving up to 99% cost reduction. Our evaluation demonstrates that influence values can be estimated with models just 0.0027% the size of full language models (we use 7B and 8B versions). We apply our algorithm of estimating influence values (called NN-CIFT: Neural Networks for effiCient Instruction Fine-Tuning) to the downstream task of subset selection for general instruction fine-tuning. In our study, we include four state-of-the-art influence functions and show no compromise in performance, despite large speedups, between NN-CIFT and the original influence functions. We provide an in-depth hyperparameter analyses of NN-CIFT. The code for our method can be found here: https://github.com/agarwalishika/NN-CIFT.

♻ ☆ Large Language Models Report Subjective Experience Under Self-Referential Processing

Cameron Berg, Diogo de Lucena, Judd Rosenblatt

Large language models sometimes produce structured, first-person descriptions that explicitly reference awareness or subjective experience. To better understand this behavior, we investigate one theoretically motivated condition under which such reports arise: self-referential processing, a computational motif emphasized across major theories of consciousness. Through a series of controlled experiments on GPT, Claude, and Gemini model families, we test whether this regime reliably shifts models toward first-person reports of subjective experience, and how such claims behave under mechanistic and behavioral probes. Four main results emerge: (1) Inducing sustained self-reference through simple prompting consistently elicits structured subjective experience reports across model families. (2) These reports are mechanistically gated by interpretable sparse-autoencoder features associated with deception and roleplay: surprisingly, suppressing deception features sharply increases the frequency of experience claims, while amplifying them minimizes such claims. (3) Structured descriptions of the self-referential state converge statistically across model families in ways not observed in any control condition. (4) The induced state yields significantly richer introspection in downstream reasoning tasks where self-reflection is only indirectly afforded. While these findings do not constitute direct evidence of consciousness, they implicate self-referential processing as a minimal and reproducible condition under which large language models generate structured first-person reports that are mechanistically gated, semantically convergent, and behaviorally generalizable. The systematic emergence of this pattern across architectures makes it a first-order scientific and ethical priority for further investigation.

♻ ☆ Let LRMs Break Free from Overthinking via Self-Braking Tuning NeurIPS 2025

Haoran Zhao, Yuchen Yan, Yongliang Shen, Haolei Xu, Wenqi Zhang, Kaitao Song, Jian Shao, Weiming Lu, Jun Xiao, Yueting Zhuang

Large reasoning models (LRMs), such as OpenAI o1 and DeepSeek-R1, have significantly enhanced their reasoning capabilities by generating longer chains of thought, demonstrating outstanding performance across a variety of tasks. However, this performance gain comes at the cost of a substantial increase in redundant reasoning during the generation process, leading to high computational overhead and exacerbating the issue of overthinking. Although numerous existing approaches aim to address the problem of overthinking, they often rely on external interventions. In this paper, we propose a novel framework, Self-Braking Tuning (SBT), which tackles overthinking from the perspective of allowing the model to regulate its own reasoning process, thus eliminating the reliance on external control mechanisms. We construct a set of overthinking identification metrics based on standard answers and design a systematic method to detect redundant reasoning. This method accurately identifies unnecessary steps within the reasoning trajectory and generates training signals for learning self-regulation behaviors. Building on this foundation, we develop a complete strategy for constructing data with adaptive reasoning lengths and introduce an innovative braking prompt mechanism that enables the model to naturally learn when to terminate reasoning at an appropriate point. Experiments across mathematical benchmarks (AIME, AMC, MATH500, GSM8K) demonstrate that our method reduces token consumption by up to 60% while maintaining comparable accuracy to unconstrained models.

comment: Accepted to NeurIPS 2025; Camera ready version, 10 pages. Github:https://github.com/ZJU-REAL/Self-Braking-Tuning Project Page: https://ZJU-REAL.github.io/SBT

♻ ☆ Model Provenance Testing for Large Language Models

Ivica Nikolic, Teodora Baluta, Prateek Saxena

Large language models are increasingly customized through fine-tuning and other adaptations, creating challenges in enforcing licensing terms and managing downstream impacts. Tracking model origins is crucial both for protecting intellectual property and for identifying derived models when biases or vulnerabilities are discovered in foundation models. We address this challenge by developing a framework for testing model provenance: Whether one model is derived from another. Our approach is based on the key observation that real-world model derivations preserve significant similarities in model outputs that can be detected through statistical analysis. Using only black-box access to models, we employ multiple hypothesis testing to compare model similarities against a baseline established by unrelated models. On two comprehensive real-world benchmarks spanning models from 30M to 4B parameters and comprising over 600 models, our tester achieves 90-95% precision and 80-90% recall in identifying derived models. These results demonstrate the viability of systematic provenance verification in production environments even when only API access is available.

♻ ☆ When Agents Trade: Live Multi-Market Trading Benchmark for LLM Agents

Lingfei Qian, Xueqing Peng, Yan Wang, Vincent Jim Zhang, Huan He, Hanley Smith, Yi Han, Yueru He, Haohang Li, Yupeng Cao, Yangyang Yu, Alejandro Lopez-Lira, Peng Lu, Jian-Yun Nie, Guojun Xiong, Jimin Huang, Sophia Ananiadou

Although Large Language Model (LLM)-based agents are increasingly used in financial trading, it remains unclear whether they can reason and adapt in live markets, as most studies test models instead of agents, cover limited periods and assets, and rely on unverified data. To address these gaps, we introduce Agent Market Arena (AMA), the first lifelong, real-time benchmark for evaluating LLM-based trading agents across multiple markets. AMA integrates verified trading data, expert-checked news, and diverse agent architectures within a unified trading framework, enabling fair and continuous comparison under real conditions. It implements four agents, including InvestorAgent as a single-agent baseline, TradeAgent and HedgeFundAgent with different risk styles, and DeepFundAgent with memory-based reasoning, and evaluates them across GPT-4o, GPT-4.1, Claude-3.5-haiku, Claude-sonnet-4, and Gemini-2.0-flash. Live experiments on both cryptocurrency and stock markets demonstrate that agent frameworks display markedly distinct behavioral patterns, spanning from aggressive risk-taking to conservative decision-making, whereas model backbones contribute less to outcome variation. AMA thus establishes a foundation for rigorous, reproducible, and continuously evolving evaluation of financial reasoning and trading intelligence in LLM-based agents.

♻ ☆ ChartMuseum: Testing Visual Reasoning Capabilities of Large Vision-Language Models NeurIPS 2025

Liyan Tang, Grace Kim, Xinyu Zhao, Thom Lake, Wenxuan Ding, Fangcong Yin, Prasann Singhal, Manya Wadhwa, Zeyu Leo Liu, Zayne Sprague, Ramya Namuduri, Bodun Hu, Juan Diego Rodriguez, Puyuan Peng, Greg Durrett

Chart understanding presents a unique challenge for large vision-language models (LVLMs), as it requires the integration of sophisticated textual and visual reasoning capabilities. However, current LVLMs exhibit a notable imbalance between these skills, falling short on visual reasoning that is difficult to perform in text. We conduct a case study using a synthetic dataset solvable only through visual reasoning and show that model performance degrades significantly with increasing visual complexity, while human performance remains robust. We then introduce ChartMuseum, a new Chart Question Answering (QA) benchmark containing 1,162 expert-annotated questions spanning multiple reasoning types, curated from real-world charts across 184 sources, specifically built to evaluate complex visual and textual reasoning. Unlike prior chart understanding benchmarks -- where frontier models perform similarly and near saturation -- our benchmark exposes a substantial gap between model and human performance, while effectively differentiating model capabilities: although humans achieve 93% accuracy, the best-performing model Gemini-2.5-Pro attains only 63.0%, and the leading open-source LVLM Qwen2.5-VL-72B-Instruct achieves only 38.5%. Moreover, on questions requiring primarily visual reasoning, all models experience a 35%-55% performance drop from text-reasoning-heavy question performance. Lastly, our qualitative error analysis reveals specific categories of visual reasoning that are challenging for current LVLMs.

comment: NeurIPS 2025 Datasets & Benchmarks

♻ ☆ Evaluating Emotion Recognition in Spoken Language Models on Emotionally Incongruent Speech ICASSP 2026

Pedro Corrêa, João Lima, Victor Moreno, Lucas Ueda, Paula Dornhofer Paro Costa

Advancements in spoken language processing have driven the development of spoken language models (SLMs), designed to achieve universal audio understanding by jointly learning text and audio representations for a wide range of tasks. Although promising results have been achieved, there is growing discussion regarding these models' generalization capabilities and the extent to which they truly integrate audio and text modalities in their internal representations. In this work, we evaluate four SLMs on the task of speech emotion recognition using a dataset of emotionally incongruent speech samples, a condition under which the semantic content of the spoken utterance conveys one emotion while speech expressiveness conveys another. Our results indicate that SLMs rely predominantly on textual semantics rather than speech emotion to perform the task, indicating that text-related representations largely dominate over acoustic representations. We release both the code and the Emotionally Incongruent Synthetic Speech dataset (EMIS) to the community.

comment: Submitted to IEEE ICASSP 2026. Copyright 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses

♻ ☆ Improving LLM Safety Alignment with Dual-Objective Optimization ICML 2025

Xuandong Zhao, Will Cai, Tianneng Shi, David Huang, Licong Lin, Song Mei, Dawn Song

Existing training-time safety alignment techniques for large language models (LLMs) remain vulnerable to jailbreak attacks. Direct preference optimization (DPO), a widely deployed alignment method, exhibits limitations in both experimental and theoretical contexts as its loss function proves suboptimal for refusal learning. Through gradient-based analysis, we identify these shortcomings and propose an improved safety alignment that disentangles DPO objectives into two components: (1) robust refusal training, which encourages refusal even when partial unsafe generations are produced, and (2) targeted unlearning of harmful knowledge. This approach significantly increases LLM robustness against a wide range of jailbreak attacks, including prefilling, suffix, and multi-turn attacks across both in-distribution and out-of-distribution scenarios. Furthermore, we introduce a method to emphasize critical refusal tokens by incorporating a reward-based token-level weighting mechanism for refusal learning, which further improves the robustness against adversarial exploits. Our research also suggests that robustness to jailbreak attacks is correlated with token distribution shifts in the training process and internal representations of refusal and harmful tokens, offering valuable directions for future research in LLM safety alignment. The code is available at https://github.com/wicai24/DOOR-Alignment

comment: ICML 2025

♻ ☆ Language Model Preference Evaluation with Multiple Weak Evaluators

Zhengyu Hu, Jieyu Zhang, Zhihan Xiong, Alexander Ratner, Kaize Ding, Ranjay Krishna

Despite the remarkable success of Large Language Models (LLMs), evaluating their outputs' quality regarding preference remains a critical challenge. While existing works usually leverage a strong LLM as the judge for comparing LLMs' response pairwisely, such a single-evaluator approach is vulnerable to cyclic preference, i.e., output A is better than B, B than C, but C is better than A, causing contradictory evaluation results. To address this, we introduce PGED (Preference Graph Ensemble and Denoise), a novel approach that leverages multiple model-based evaluators to construct preference graphs, and then ensembles and denoises these graphs for acyclic, non-contradictory evaluation results. We provide theoretical guarantees for our framework, demonstrating its efficacy in recovering the ground truth preference structure. Extensive experiments on ten benchmarks demonstrate PGED 's superiority in three applications: 1) model ranking for evaluation, 2) response selection for test-time scaling, and 3) data selection for model fine-tuning. Notably, PGED combines small LLM evaluators (e.g., Llama3-8B, Mistral-7B, Qwen2-7B) to outperform strong ones (e.g., Qwen2-72B), showcasing its effectiveness in enhancing evaluation reliability and improving model performance.

☆ Are Video Models Ready as Zero-Shot Reasoners? An Empirical Study with the MME-CoF Benchmark

Ziyu Guo, Xinyan Chen, Renrui Zhang, Ruichuan An, Yu Qi, Dongzhi Jiang, Xiangtai Li, Manyuan Zhang, Hongsheng Li, Pheng-Ann Heng

Recent video generation models can produce high-fidelity, temporally coherent videos, indicating that they may encode substantial world knowledge. Beyond realistic synthesis, they also exhibit emerging behaviors indicative of visual perception, modeling, and manipulation. Yet, an important question still remains: Are video models ready to serve as zero-shot reasoners in challenging visual reasoning scenarios? In this work, we conduct an empirical study to comprehensively investigate this question, focusing on the leading and popular Veo-3. We evaluate its reasoning behavior across 12 dimensions, including spatial, geometric, physical, temporal, and embodied logic, systematically characterizing both its strengths and failure modes. To standardize this study, we curate the evaluation data into MME-CoF, a compact benchmark that enables in-depth and thorough assessment of Chain-of-Frame (CoF) reasoning. Our findings reveal that while current video models demonstrate promising reasoning patterns on short-horizon spatial coherence, fine-grained grounding, and locally consistent dynamics, they remain limited in long-horizon causal reasoning, strict geometric constraints, and abstract logic. Overall, they are not yet reliable as standalone zero-shot reasoners, but exhibit encouraging signs as complementary visual engines alongside dedicated reasoning models. Project page: https://video-cof.github.io

comment: Project Page: https://video-cof.github.io

☆ OmniX: From Unified Panoramic Generation and Perception to Graphics-Ready 3D Scenes

Yukun Huang, Jiwen Yu, Yanning Zhou, Jianan Wang, Xintao Wang, Pengfei Wan, Xihui Liu

There are two prevalent ways to constructing 3D scenes: procedural generation and 2D lifting. Among them, panorama-based 2D lifting has emerged as a promising technique, leveraging powerful 2D generative priors to produce immersive, realistic, and diverse 3D environments. In this work, we advance this technique to generate graphics-ready 3D scenes suitable for physically based rendering (PBR), relighting, and simulation. Our key insight is to repurpose 2D generative models for panoramic perception of geometry, textures, and PBR materials. Unlike existing 2D lifting approaches that emphasize appearance generation and ignore the perception of intrinsic properties, we present OmniX, a versatile and unified framework. Based on a lightweight and efficient cross-modal adapter structure, OmniX reuses 2D generative priors for a broad range of panoramic vision tasks, including panoramic perception, generation, and completion. Furthermore, we construct a large-scale synthetic panorama dataset containing high-quality multimodal panoramas from diverse indoor and outdoor scenes. Extensive experiments demonstrate the effectiveness of our model in panoramic visual perception and graphics-ready 3D scene generation, opening new possibilities for immersive and physically realistic virtual world generation.

comment: Project page: https://yukun-huang.github.io/OmniX/

☆ Masked Diffusion Captioning for Visual Feature Learning EMNLP 2025

Chao Feng, Zihao Wei, Andrew Owens

We learn visual features by captioning images with an image-conditioned masked diffusion language model, a formulation we call masked diffusion captioning (MDC). During training, text tokens in each image-caption pair are masked at a randomly chosen ratio, and a decoder conditioned on visual features is trained to reconstruct the original text. After training, the learned visual features can be applied to downstream vision tasks. Unlike autoregressive captioning, the strength of the visual learning signal in MDC does not depend on each token's position in the sequence, reducing the need for auxiliary objectives. Linear probing experiments across a variety of academic-scale models and datasets show that the learned visual features are competitive with those produced by autoregressive and contrastive approaches.

comment: EMNLP 2025 (Findings). Project page: https://cfeng16.github.io/mdlm4vfl/

☆ SEE4D: Pose-Free 4D Generation via Auto-Regressive Video Inpainting

Dongyue Lu, Ao Liang, Tianxin Huang, Xiao Fu, Yuyang Zhao, Baorui Ma, Liang Pan, Wei Yin, Lingdong Kong, Wei Tsang Ooi, Ziwei Liu

Immersive applications call for synthesizing spatiotemporal 4D content from casual videos without costly 3D supervision. Existing video-to-4D methods typically rely on manually annotated camera poses, which are labor-intensive and brittle for in-the-wild footage. Recent warp-then-inpaint approaches mitigate the need for pose labels by warping input frames along a novel camera trajectory and using an inpainting model to fill missing regions, thereby depicting the 4D scene from diverse viewpoints. However, this trajectory-to-trajectory formulation often entangles camera motion with scene dynamics and complicates both modeling and inference. We introduce SEE4D, a pose-free, trajectory-to-camera framework that replaces explicit trajectory prediction with rendering to a bank of fixed virtual cameras, thereby separating camera control from scene modeling. A view-conditional video inpainting model is trained to learn a robust geometry prior by denoising realistically synthesized warped images and to inpaint occluded or missing regions across virtual viewpoints, eliminating the need for explicit 3D annotations. Building on this inpainting core, we design a spatiotemporal autoregressive inference pipeline that traverses virtual-camera splines and extends videos with overlapping windows, enabling coherent generation at bounded per-step complexity. We validate See4D on cross-view video generation and sparse reconstruction benchmarks. Across quantitative metrics and qualitative assessments, our method achieves superior generalization and improved performance relative to pose- or trajectory-conditioned baselines, advancing practical 4D world modeling from casual videos.

comment: 26 pages; 21 figures; 3 tables; project page: https://see-4d.github.io/

☆ Scaling Image Geo-Localization to Continent Level NeurIPS 2025

Philipp Lindenberger, Paul-Edouard Sarlin, Jan Hosang, Matteo Balice, Marc Pollefeys, Simon Lynen, Eduard Trulls

Determining the precise geographic location of an image at a global scale remains an unsolved challenge. Standard image retrieval techniques are inefficient due to the sheer volume of images (>100M) and fail when coverage is insufficient. Scalable solutions, however, involve a trade-off: global classification typically yields coarse results (10+ kilometers), while cross-view retrieval between ground and aerial imagery suffers from a domain gap and has been primarily studied on smaller regions. This paper introduces a hybrid approach that achieves fine-grained geo-localization across a large geographic expanse the size of a continent. We leverage a proxy classification task during training to learn rich feature representations that implicitly encode precise location information. We combine these learned prototypes with embeddings of aerial imagery to increase robustness to the sparsity of ground-level data. This enables direct, fine-grained retrieval over areas spanning multiple countries. Our extensive evaluation demonstrates that our approach can localize within 200m more than 68\% of queries of a dataset covering a large part of Europe. The code is publicly available at https://scaling-geoloc.github.io.

comment: NeurIPS 2025

☆ The Quest for Generalizable Motion Generation: Data, Model, and Evaluation

Jing Lin, Ruisi Wang, Junzhe Lu, Ziqi Huang, Guorui Song, Ailing Zeng, Xian Liu, Chen Wei, Wanqi Yin, Qingping Sun, Zhongang Cai, Lei Yang, Ziwei Liu

Despite recent advances in 3D human motion generation (MoGen) on standard benchmarks, existing models still face a fundamental bottleneck in their generalization capability. In contrast, adjacent generative fields, most notably video generation (ViGen), have demonstrated remarkable generalization in modeling human behaviors, highlighting transferable insights that MoGen can leverage. Motivated by this observation, we present a comprehensive framework that systematically transfers knowledge from ViGen to MoGen across three key pillars: data, modeling, and evaluation. First, we introduce ViMoGen-228K, a large-scale dataset comprising 228,000 high-quality motion samples that integrates high-fidelity optical MoCap data with semantically annotated motions from web videos and synthesized samples generated by state-of-the-art ViGen models. The dataset includes both text-motion pairs and text-video-motion triplets, substantially expanding semantic diversity. Second, we propose ViMoGen, a flow-matching-based diffusion transformer that unifies priors from MoCap data and ViGen models through gated multimodal conditioning. To enhance efficiency, we further develop ViMoGen-light, a distilled variant that eliminates video generation dependencies while preserving strong generalization. Finally, we present MBench, a hierarchical benchmark designed for fine-grained evaluation across motion quality, prompt fidelity, and generalization ability. Extensive experiments show that our framework significantly outperforms existing approaches in both automatic and human evaluations. The code, data, and benchmark will be made publicly available.

☆ HEIR: Learning Graph-Based Motion Hierarchies

Cheng Zheng, William Koch, Baiang Li, Felix Heide

Hierarchical structures of motion exist across research fields, including computer vision, graphics, and robotics, where complex dynamics typically arise from coordinated interactions among simpler motion components. Existing methods to model such dynamics typically rely on manually-defined or heuristic hierarchies with fixed motion primitives, limiting their generalizability across different tasks. In this work, we propose a general hierarchical motion modeling method that learns structured, interpretable motion relationships directly from data. Our method represents observed motions using graph-based hierarchies, explicitly decomposing global absolute motions into parent-inherited patterns and local motion residuals. We formulate hierarchy inference as a differentiable graph learning problem, where vertices represent elemental motions and directed edges capture learned parent-child dependencies through graph neural networks. We evaluate our hierarchical reconstruction approach on three examples: 1D translational motion, 2D rotational motion, and dynamic 3D scene deformation via Gaussian splatting. Experimental results show that our method reconstructs the intrinsic motion hierarchy in 1D and 2D cases, and produces more realistic and interpretable deformations compared to the baseline on dynamic 3D Gaussian splatting scenes. By providing an adaptable, data-driven hierarchical modeling paradigm, our method offers a formulation applicable to a broad range of motion-centric tasks. Project Page: https://light.princeton.edu/HEIR/

comment: Code link: https://github.com/princeton-computational-imaging/HEIR

☆ Clone Deterministic 3D Worlds with Geometrically-Regularized World Models

Zaishuo Xia, Yukuan Lu, Xinyi Li, Yifan Xu, Yubei Chen

A world model is an internal model that simulates how the world evolves. Given past observations and actions, it predicts the future of both the embodied agent and its environment. Accurate world models are essential for enabling agents to think, plan, and reason effectively in complex, dynamic settings. Despite rapid progress, current world models remain brittle and degrade over long horizons. We argue that a central cause is representation quality: exteroceptive inputs (e.g., images) are high-dimensional, and lossy or entangled latents make dynamics learning unnecessarily hard. We therefore ask whether improving representation learning alone can substantially improve world-model performance. In this work, we take a step toward building a truly accurate world model by addressing a fundamental yet open problem: constructing a model that can fully clone and overfit to a deterministic 3D world. We propose Geometrically-Regularized World Models (GRWM), which enforces that consecutive points along a natural sensory trajectory remain close in latent representation space. This approach yields significantly improved latent representations that align closely with the true topology of the environment. GRWM is plug-and-play, requires only minimal architectural modification, scales with trajectory length, and is compatible with diverse latent generative backbones. Across deterministic 3D settings and long-horizon prediction tasks, GRWM significantly increases rollout fidelity and stability. Analyses show that its benefits stem from learning a latent manifold with superior geometric structure. These findings support a clear takeaway: improving representation learning is a direct and useful path to robust world models, delivering reliable long-horizon predictions without enlarging the dynamics module.

☆ ChartAB: A Benchmark for Chart Grounding & Dense Alignment

Aniruddh Bansal, Davit Soselia, Dang Nguyen, Tianyi Zhou

Charts play an important role in visualization, reasoning, data analysis, and the exchange of ideas among humans. However, existing vision-language models (VLMs) still lack accurate perception of details and struggle to extract fine-grained structures from charts. Such limitations in chart grounding also hinder their ability to compare multiple charts and reason over them. In this paper, we introduce a novel "ChartAlign Benchmark (ChartAB)" to provide a comprehensive evaluation of VLMs in chart grounding tasks, i.e., extracting tabular data, localizing visualization elements, and recognizing various attributes from charts of diverse types and complexities. We design a JSON template to facilitate the calculation of evaluation metrics specifically tailored for each grounding task. By incorporating a novel two-stage inference workflow, the benchmark can further evaluate VLMs' capability to align and compare elements/attributes across two charts. Our analysis of evaluations on several recent VLMs reveals new insights into their perception biases, weaknesses, robustness, and hallucinations in chart understanding. These findings highlight the fine-grained discrepancies among VLMs in chart understanding tasks and point to specific skills that need to be strengthened in current models.

☆ Surpassing state of the art on AMD area estimation from RGB fundus images through careful selection of U-Net architectures and loss functions for class imbalance

Valentyna Starodub, Mantas Lukoševičius

Age-related macular degeneration (AMD) is one of the leading causes of irreversible vision impairment in people over the age of 60. This research focuses on semantic segmentation for AMD lesion detection in RGB fundus images, a non-invasive and cost-effective imaging technique. The results of the ADAM challenge - the most comprehensive AMD detection from RGB fundus images research competition and open dataset to date - serve as a benchmark for our evaluation. Taking the U-Net connectivity as a base of our framework, we evaluate and compare several approaches to improve the segmentation model's architecture and training pipeline, including pre-processing techniques, encoder (backbone) deep network types of varying complexity, and specialized loss functions to mitigate class imbalances on image and pixel levels. The main outcome of this research is the final configuration of the AMD detection framework, which outperforms all the prior ADAM challenge submissions on the multi-class segmentation of different AMD lesion types in non-invasive RGB fundus images. The source code used to conduct the experiments presented in this paper is made freely available.

☆ SteerVLM: Robust Model Control through Lightweight Activation Steering for Vision Language Models

Anushka Sivakumar, Andrew Zhang, Zaber Hakim, Chris Thomas

This work introduces SteerVLM, a lightweight steering module designed to guide Vision-Language Models (VLMs) towards outputs that better adhere to desired instructions. Our approach learns from the latent embeddings of paired prompts encoding target and converse behaviors to dynamically adjust activations connecting the language modality with image context. This allows for fine-grained, inference-time control over complex output semantics without modifying model weights while preserving performance on off-target tasks. Our steering module requires learning parameters equal to 0.14% of the original VLM's size. Our steering module gains model control through dimension-wise activation modulation and adaptive steering across layers without requiring pre-extracted static vectors or manual tuning of intervention points. Furthermore, we introduce VNIA (Visual Narrative Intent Alignment), a multimodal dataset specifically created to facilitate the development and evaluation of VLM steering techniques. Our method outperforms existing intervention techniques on steering and hallucination mitigation benchmarks for VLMs and proposes a robust solution for multimodal model control through activation engineering.

☆ MORE: Multi-Organ Medical Image REconstruction Dataset

Shaokai Wu, Yapan Guo, Yanbiao Ji, Jing Tong, Yuxiang Lu, Mei Li, Suizhi Huang, Yue Ding, Hongtao Lu

CT reconstruction provides radiologists with images for diagnosis and treatment, yet current deep learning methods are typically limited to specific anatomies and datasets, hindering generalization ability to unseen anatomies and lesions. To address this, we introduce the Multi-Organ medical image REconstruction (MORE) dataset, comprising CT scans across 9 diverse anatomies with 15 lesion types. This dataset serves two key purposes: (1) enabling robust training of deep learning models on extensive, heterogeneous data, and (2) facilitating rigorous evaluation of model generalization for CT reconstruction. We further establish a strong baseline solution that outperforms prior approaches under these challenging conditions. Our results demonstrate that: (1) a comprehensive dataset helps improve the generalization capability of models, and (2) optimization-based methods offer enhanced robustness for unseen anatomies. The MORE dataset is freely accessible under CC-BY-NC 4.0 at our project page https://more-med.github.io/

comment: Accepted to ACMMM 2025

☆ ProstNFound+: A Prospective Study using Medical Foundation Models for Prostate Cancer Detection

Paul F. R. Wilson, Mohamed Harmanani, Minh Nguyen Nhat To, Amoon Jamzad, Tarek Elghareb, Zhuoxin Guo, Adam Kinnaird, Brian Wodlinger, Purang Abolmaesumi, Parvin Mousavi

Purpose: Medical foundation models (FMs) offer a path to build high-performance diagnostic systems. However, their application to prostate cancer (PCa) detection from micro-ultrasound ({\mu}US) remains untested in clinical settings. We present ProstNFound+, an adaptation of FMs for PCa detection from {\mu}US, along with its first prospective validation. Methods: ProstNFound+ incorporates a medical FM, adapter tuning, and a custom prompt encoder that embeds PCa-specific clinical biomarkers. The model generates a cancer heatmap and a risk score for clinically significant PCa. Following training on multi-center retrospective data, the model is prospectively evaluated on data acquired five years later from a new clinical site. Model predictions are benchmarked against standard clinical scoring protocols (PRI-MUS and PI-RADS). Results: ProstNFound+ shows strong generalization to the prospective data, with no performance degradation compared to retrospective evaluation. It aligns closely with clinical scores and produces interpretable heatmaps consistent with biopsy-confirmed lesions. Conclusion: The results highlight its potential for clinical deployment, offering a scalable and interpretable alternative to expert-driven protocols.

☆ The Impact and Outlook of 3D Gaussian Splatting

Bernhard Kerbl

Since its introduction, 3D Gaussian Splatting (3DGS) has rapidly transformed the landscape of 3D scene representations, inspiring an extensive body of associated research. Follow-up work includes analyses and contributions that enhance the efficiency, scalability, and real-world applicability of 3DGS. In this summary, we present an overview of several key directions that have emerged in the wake of 3DGS. We highlight advances enabling resource-efficient training and rendering, the evolution toward dynamic (or four-dimensional, 4DGS) representations, and deeper exploration of the mathematical foundations underlying its appearance modeling and rendering process. Furthermore, we examine efforts to bring 3DGS to mobile and virtual reality platforms, its extension to massive-scale environments, and recent progress toward near-instant radiance field reconstruction via feed-forward or distributed computation. Collectively, these developments illustrate how 3DGS has evolved from a breakthrough representation into a versatile and foundational tool for 3D vision and graphics.

comment: Article written for Frontiers of Science Award, International Congress on Basic Science, 2025

☆ Process Integrated Computer Vision for Real-Time Failure Prediction in Steel Rolling Mill

Vaibhav Kurrey, Sivakalyan Pujari, Gagan Raj Gupta

We present a long-term deployment study of a machine vision-based anomaly detection system for failure prediction in a steel rolling mill. The system integrates industrial cameras to monitor equipment operation, alignment, and hot bar motion in real time along the process line. Live video streams are processed on a centralized video server using deep learning models, enabling early prediction of equipment failures and process interruptions, thereby reducing unplanned breakdown costs. Server-based inference minimizes the computational load on industrial process control systems (PLCs), supporting scalable deployment across production lines with minimal additional resources. By jointly analyzing sensor data from data acquisition systems and visual inputs, the system identifies the location and probable root causes of failures, providing actionable insights for proactive maintenance. This integrated approach enhances operational reliability, productivity, and profitability in industrial manufacturing environments.

☆ Improving Classification of Occluded Objects through Scene Context

Courtney M. King, Daniel D. Leeds, Damian Lyons, George Kalaitzis

The presence of occlusions has provided substantial challenges to typically-powerful object recognition algorithms. Additional sources of information can be extremely valuable to reduce errors caused by occlusions. Scene context is known to aid in object recognition in biological vision. In this work, we attempt to add robustness into existing Region Proposal Network-Deep Convolutional Neural Network (RPN-DCNN) object detection networks through two distinct scene-based information fusion techniques. We present one algorithm under each methodology: the first operates prior to prediction, selecting a custom object network to use based on the identified background scene, and the second operates after detection, fusing scene knowledge into initial object scores output by the RPN. We demonstrate our algorithms on challenging datasets featuring partial occlusions, which show overall improvement in both recall and precision against baseline methods. In addition, our experiments contrast multiple training methodologies for occlusion handling, finding that training on a combination of both occluded and unoccluded images demonstrates an improvement over the others. Our method is interpretable and can easily be adapted to other datasets, offering many future directions for research and practical applications.

☆ BRIQA: Balanced Reweighting in Image Quality Assessment of Pediatric Brain MRI

Alya Almsouti, Ainur Khamitova, Darya Taratynova, Mohammad Yaqub

Assessing the severity of artifacts in pediatric brain Magnetic Resonance Imaging (MRI) is critical for diagnostic accuracy, especially in low-field systems where the signal-to-noise ratio is reduced. Manual quality assessment is time-consuming and subjective, motivating the need for robust automated solutions. In this work, we propose BRIQA (Balanced Reweighting in Image Quality Assessment), which addresses class imbalance in artifact severity levels. BRIQA uses gradient-based loss reweighting to dynamically adjust per-class contributions and employs a rotating batching scheme to ensure consistent exposure to underrepresented classes. Through experiments, no single architecture performs best across all artifact types, emphasizing the importance of architectural diversity. The rotating batching configuration improves performance across metrics by promoting balanced learning when combined with cross-entropy loss. BRIQA improves average macro F1 score from 0.659 to 0.706, with notable gains in Noise (0.430), Zipper (0.098), Positioning (0.097), Contrast (0.217), Motion (0.022), and Banding (0.012) artifact severity classification. The code is available at https://github.com/BioMedIA-MBZUAI/BRIQA.

☆ Towards Reliable Sea Ice Drift Estimation in the Arctic Deep Learning Optical Flow on RADARSAT-2

Daniela Martin, Joseph Gallego

Accurate estimation of sea ice drift is critical for Arctic navigation, climate research, and operational forecasting. While optical flow, a computer vision technique for estimating pixel wise motion between consecutive images, has advanced rapidly in computer vision, its applicability to geophysical problems and to satellite SAR imagery remains underexplored. Classical optical flow methods rely on mathematical models and strong assumptions about motion, which limit their accuracy in complex scenarios. Recent deep learning based approaches have substantially improved performance and are now the standard in computer vision, motivating their application to sea ice drift estimation. We present the first large scale benchmark of 48 deep learning optical flow models on RADARSAT 2 ScanSAR sea ice imagery, evaluated with endpoint error (EPE) and Fl all metrics against GNSS tracked buoys. Several models achieve sub kilometer accuracy (EPE 6 to 8 pixels, 300 to 400 m), a small error relative to the spatial scales of sea ice motion and typical navigation requirements in the Arctic. Our results demonstrate that the models are capable of capturing consistent regional drift patterns and that recent deep learning based optical flow methods, which have substantially improved motion estimation accuracy compared to classical methods, can be effectively transferred to polar remote sensing. Optical flow produces spatially continuous drift fields, providing motion estimates for every image pixel rather than at sparse buoy locations, offering new opportunities for navigation and climate modeling.

☆ All You Need for Object Detection: From Pixels, Points, and Prompts to Next-Gen Fusion and Multimodal LLMs/VLMs in Autonomous Vehicles

Sayed Pedram Haeri Boroujeni, Niloufar Mehrabi, Hazim Alzorgan, Ahmad Sarlak, Mahlagha Fazeli, Abolfazl Razi

Autonomous Vehicles (AVs) are transforming the future of transportation through advances in intelligent perception, decision-making, and control systems. However, their success is tied to one core capability, reliable object detection in complex and multimodal environments. While recent breakthroughs in Computer Vision (CV) and Artificial Intelligence (AI) have driven remarkable progress, the field still faces a critical challenge as knowledge remains fragmented across multimodal perception, contextual reasoning, and cooperative intelligence. This survey bridges that gap by delivering a forward-looking analysis of object detection in AVs, emphasizing emerging paradigms such as Vision-Language Models (VLMs), Large Language Models (LLMs), and Generative AI rather than re-examining outdated techniques. We begin by systematically reviewing the fundamental spectrum of AV sensors (camera, ultrasonic, LiDAR, and Radar) and their fusion strategies, highlighting not only their capabilities and limitations in dynamic driving environments but also their potential to integrate with recent advances in LLM/VLM-driven perception frameworks. Next, we introduce a structured categorization of AV datasets that moves beyond simple collections, positioning ego-vehicle, infrastructure-based, and cooperative datasets (e.g., V2V, V2I, V2X, I2I), followed by a cross-analysis of data structures and characteristics. Ultimately, we analyze cutting-edge detection methodologies, ranging from 2D and 3D pipelines to hybrid sensor fusion, with particular attention to emerging transformer-driven approaches powered by Vision Transformers (ViTs), Large and Small Language Models (SLMs), and VLMs. By synthesizing these perspectives, our survey delivers a clear roadmap of current capabilities, open challenges, and future opportunities.

☆ SAMRI: Segment Anything Model for MRI

Zhao Wang, Wei Dai, Thuy Thanh Dao, Steffen Bollmann, Hongfu Sun, Craig Engstrom, Shekhar S. Chandra

Accurate magnetic resonance imaging (MRI) segmentation is crucial for clinical decision-making, but remains labor-intensive when performed manually. Convolutional neural network (CNN)-based methods can be accurate and efficient, but often generalize poorly to MRI's variable contrast, intensity inhomogeneity, and protocols. Although the transformer-based Segment Anything Model (SAM) has demonstrated remarkable generalizability in natural images, existing adaptations often treat MRI as another imaging modality, overlooking these modality-specific challenges. We present SAMRI, an MRI-specialized SAM trained and validated on 1.1 million labeled MR slices spanning whole-body organs and pathologies. We demonstrate that SAM can be effectively adapted to MRI by simply fine-tuning its mask decoder using a two-stage strategy, reducing training time by 94% and trainable parameters by 96% versus full-model retraining. Across diverse MRI segmentation tasks, SAMRI achieves a mean Dice of 0.87, delivering state-of-the-art accuracy across anatomical regions and robust generalization on unseen structures, particularly small and clinically important structures.

☆ PT-DETR: Small Target Detection Based on Partially-Aware Detail Focus

Bingcong Huo, Zhiming Wang

To address the challenges in UAV object detection, such as complex backgrounds, severe occlusion, dense small objects, and varying lighting conditions,this paper proposes PT-DETR based on RT-DETR, a novel detection algorithm specifically designed for small objects in UAV imagery. In the backbone network, we introduce the Partially-Aware Detail Focus (PADF) Module to enhance feature extraction for small objects. Additionally,we design the Median-Frequency Feature Fusion (MFFF) module,which effectively improves the model's ability to capture small-object details and contextual information. Furthermore,we incorporate Focaler-SIoU to strengthen the model's bounding box matching capability and increase its sensitivity to small-object features, thereby further enhancing detection accuracy and robustness. Compared with RT-DETR, our PT-DETR achieves mAP improvements of 1.6% and 1.7% on the VisDrone2019 dataset with lower computational complexity and fewer parameters, demonstrating its robustness and feasibility for small-object detection tasks.

☆ Spiking Patches: Asynchronous, Sparse, and Efficient Tokens for Event Cameras

Christoffer Koo Øhrstrøm, Ronja Güldenring, Lazaros Nalpantidis

We propose tokenization of events and present a tokenizer, Spiking Patches, specifically designed for event cameras. Given a stream of asynchronous and spatially sparse events, our goal is to discover an event representation that preserves these properties. Prior works have represented events as frames or as voxels. However, while these representations yield high accuracy, both frames and voxels are synchronous and decrease the spatial sparsity. Spiking Patches gives the means to preserve the unique properties of event cameras and we show in our experiments that this comes without sacrificing accuracy. We evaluate our tokenizer using a GNN, PCN, and a Transformer on gesture recognition and object detection. Tokens from Spiking Patches yield inference times that are up to 3.4x faster than voxel-based tokens and up to 10.4x faster than frames. We achieve this while matching their accuracy and even surpassing in some cases with absolute improvements up to 3.8 for gesture recognition and up to 1.4 for object detection. Thus, tokenization constitutes a novel direction in event-based vision and marks a step towards methods that preserve the properties of event cameras.

☆ CYPRESS: Crop Yield Prediction via Regression on Prithvi's Encoder for Satellite Sensing

Shayan Nejadshamsi, Yuanyuan Zhang, Shadi Zaki, Brock Porth, Lysa Porth, Vahab Khoshdel

Accurate and timely crop yield prediction is crucial for global food security and modern agricultural management. Traditional methods often lack the scalability and granularity required for precision farming. This paper introduces CYPRESS (Crop Yield Prediction via Regression on Prithvi's Encoder for Satellite Sensing), a deep learning model designed for high-resolution, intra-field canola yield prediction. CYPRESS leverages a pre-trained, large-scale geospatial foundation model (Prithvi-EO-2.0-600M) and adapts it for a continuous regression task, transforming multi-temporal satellite imagery into dense, pixel-level yield maps. Evaluated on a comprehensive dataset from the Canadian Prairies, CYPRESS demonstrates superior performance over existing deep learning-based yield prediction models, highlighting the effectiveness of fine-tuning foundation models for specialized agricultural applications. By providing a continuous, high-resolution output, CYPRESS offers a more actionable tool for precision agriculture than conventional classification or county-level aggregation methods. This work validates a novel approach that bridges the gap between large-scale Earth observation and on-farm decision-making, offering a scalable solution for detailed agricultural monitoring.

☆ ResMatching: Noise-Resilient Computational Super-Resolution via Guided Conditional Flow Matching

Anirban Ray, Vera Galinova, Florian Jug

Computational Super-Resolution (CSR) in fluorescence microscopy has, despite being an ill-posed problem, a long history. At its very core, CSR is about finding a prior that can be used to extrapolate frequencies in a micrograph that have never been imaged by the image-generating microscope. It stands to reason that, with the advent of better data-driven machine learning techniques, stronger prior can be learned and hence CSR can lead to better results. Here, we present ResMatching, a novel CSR method that uses guided conditional flow matching to learn such improved data-priors. We evaluate ResMatching on 4 diverse biological structures from the BioSR dataset and compare its results against 7 baselines. ResMatching consistently achieves competitive results, demonstrating in all cases the best trade-off between data fidelity and perceptual realism. We observe that CSR using ResMatching is particularly effective in cases where a strong prior is hard to learn, e.g. when the given low-resolution images contain a lot of noise. Additionally, we show that ResMatching can be used to sample from an implicitly learned posterior distribution and that this distribution is calibrated for all tested use-cases, enabling our method to deliver a pixel-wise data-uncertainty term that can guide future users to reject uncertain predictions.

comment: 5 pages, 4 figures

☆ Emu3.5: Native Multimodal Models are World Learners

Yufeng Cui, Honghao Chen, Haoge Deng, Xu Huang, Xinghang Li, Jirong Liu, Yang Liu, Zhuoyan Luo, Jinsheng Wang, Wenxuan Wang, Yueze Wang, Chengyuan Wang, Fan Zhang, Yingli Zhao, Ting Pan, Xianduo Li, Zecheng Hao, Wenxuan Ma, Zhuo Chen, Yulong Ao, Tiejun Huang, Zhongyuan Wang, Xinlong Wang

We introduce Emu3.5, a large-scale multimodal world model that natively predicts the next state across vision and language. Emu3.5 is pre-trained end-to-end with a unified next-token prediction objective on a corpus of vision-language interleaved data containing over 10 trillion tokens, primarily derived from sequential frames and transcripts of internet videos. The model naturally accepts interleaved vision-language inputs and generates interleaved vision-language outputs. Emu3.5 is further post-trained with large-scale reinforcement learning to enhance multimodal reasoning and generation. To improve inference efficiency, we propose Discrete Diffusion Adaptation (DiDA), which converts token-by-token decoding into bidirectional parallel prediction, accelerating per-image inference by about 20x without sacrificing performance. Emu3.5 exhibits strong native multimodal capabilities, including long-horizon vision-language generation, any-to-image (X2I) generation, and complex text-rich image generation. It also exhibits generalizable world-modeling abilities, enabling spatiotemporally consistent world exploration and open-world embodied manipulation across diverse scenarios and tasks. For comparison, Emu3.5 achieves performance comparable to Gemini 2.5 Flash Image (Nano Banana) on image generation and editing tasks and demonstrates superior results on a suite of interleaved generation tasks. We open-source Emu3.5 at https://github.com/baaivision/Emu3.5 to support community research.

comment: project page: https://emu.world

☆ CATCH: A Modular Cross-domain Adaptive Template with Hook

Xinjin Li, Yulie Lu, Jinghan Cao, Yu Ma, Zhenglin Li, Yeyang Zhou

Recent advances in Visual Question Answering (VQA) have demonstrated impressive performance in natural image domains, with models like LLaVA leveraging large language models (LLMs) for open-ended reasoning. However, their generalization degrades significantly when transferred to out-of-domain scenarios such as remote sensing, medical imaging, or math diagrams, due to large distributional shifts and the lack of effective domain adaptation mechanisms. Existing approaches typically rely on per-domain fine-tuning or bespoke pipelines, which are costly, inflexible, and not scalable across diverse tasks. In this paper, we propose CATCH, a plug-and-play framework for cross-domain adaptation that improves the generalization of VQA models while requiring minimal changes to their core architecture. Our key idea is to decouple visual and linguistic adaptation by introducing two lightweight modules: a domain classifier to identify the input image type, and a dual adapter mechanism comprising a Prompt Adapter for language modulation and a Visual Adapter for vision feature adjustment. Both modules are dynamically injected via a unified hook interface, requiring no retraining of the backbone model. Experimental results across four domain-specific VQA benchmarks demonstrate that our framework achieves consistent performance gains without retraining the backbone model, including +2.3 BLEU on MathVQA, +2.6 VQA on MedVQA-RAD, and +3.1 ROUGE on ChartQA. These results highlight that CATCH provides a scalable and extensible approach to multi-domain VQA, enabling practical deployment across diverse application domains.

☆ Dynamic Context-Aware Scene Reasoning Using Vision-Language Alignment in Zero-Shot Real-World Scenarios

Manjunath Prasad Holenarasipura Rajiv, B. M. Vidyavathi

In real-world environments, AI systems often face unfamiliar scenarios without labeled data, creating a major challenge for conventional scene understanding models. The inability to generalize across unseen contexts limits the deployment of vision-based applications in dynamic, unstructured settings. This work introduces a Dynamic Context-Aware Scene Reasoning framework that leverages Vision-Language Alignment to address zero-shot real-world scenarios. The goal is to enable intelligent systems to infer and adapt to new environments without prior task-specific training. The proposed approach integrates pre-trained vision transformers and large language models to align visual semantics with natural language descriptions, enhancing contextual comprehension. A dynamic reasoning module refines predictions by combining global scene cues and object-level interactions guided by linguistic priors. Extensive experiments on zero-shot benchmarks such as COCO, Visual Genome, and Open Images demonstrate up to 18% improvement in scene understanding accuracy over baseline models in complex and unseen environments. Results also show robust performance in ambiguous or cluttered scenes due to the synergistic fusion of vision and language. This framework offers a scalable and interpretable approach for context-aware reasoning, advancing zero-shot generalization in dynamic real-world settings.

comment: Preprint under review at IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2025

☆ Comparative Analysis of Deep Learning Models for Olive Tree Crown and Shadow Segmentation Towards Biovolume Estimation

Wondimagegn Abebe Demissie, Stefano Roccella, Rudy Rossetto, Antonio Minnocci, Andrea Vannini, Luca Sebastiani

Olive tree biovolume estimation is a key task in precision agriculture, supporting yield prediction and resource management, especially in Mediterranean regions severely impacted by climate-induced stress. This study presents a comparative analysis of three deep learning models U-Net, YOLOv11m-seg, and Mask RCNN for segmenting olive tree crowns and their shadows in ultra-high resolution UAV imagery. The UAV dataset, acquired over Vicopisano, Italy, includes manually annotated crown and shadow masks. Building on these annotations, the methodology emphasizes spatial feature extraction and robust segmentation; per-tree biovolume is then estimated by combining crown projected area with shadow-derived height using solar geometry. In testing, Mask R-CNN achieved the best overall accuracy (F1 = 0.86; mIoU = 0.72), while YOLOv11m-seg provided the fastest throughput (0.12 second per image). The estimated biovolumes spanned from approximately 4 to 24 cubic meters, reflecting clear structural differences among trees. These results indicate Mask R-CNN is preferable when biovolume accuracy is paramount, whereas YOLOv11m-seg suits large-area deployments where speed is critical; U-Net remains a lightweight, high-sensitivity option. The framework enables accurate, scalable orchard monitoring and can be further strengthened with DEM or DSM integration and field calibration for operational decision support.

comment: 6 pages, 2025 IEEE International Workshop on Metrology for Agriculture and Forestry (MetroAgriFor)

☆ AdSum: Two-stream Audio-visual Summarization for Automated Video Advertisement Clipping

Wen Xie, Yanjun Zhu, Gijs Overgoor, Yakov Bart, Agata Lapedriza Garcia, Sarah Ostadabbas

Advertisers commonly need multiple versions of the same advertisement (ad) at varying durations for a single campaign. The traditional approach involves manually selecting and re-editing shots from longer video ads to create shorter versions, which is labor-intensive and time-consuming. In this paper, we introduce a framework for automated video ad clipping using video summarization techniques. We are the first to frame video clipping as a shot selection problem, tailored specifically for advertising. Unlike existing general video summarization methods that primarily focus on visual content, our approach emphasizes the critical role of audio in advertising. To achieve this, we develop a two-stream audio-visual fusion model that predicts the importance of video frames, where importance is defined as the likelihood of a frame being selected in the firm-produced short ad. To address the lack of ad-specific datasets, we present AdSum204, a novel dataset comprising 102 pairs of 30-second and 15-second ads from real advertising campaigns. Extensive experiments demonstrate that our model outperforms state-of-the-art methods across various metrics, including Average Precision, Area Under Curve, Spearman, and Kendall.

comment: Accepted at 32nd International Conference on MultiMedia Modeling

☆ SA$^{2}$Net: Scale-Adaptive Structure-Affinity Transformation for Spine Segmentation from Ultrasound Volume Projection Imaging

Hao Xie, Zixun Huang, Yushen Zuo, Yakun Ju, Frank H. F. Leung, N. F. Law, Kin-Man Lam, Yong-Ping Zheng, Sai Ho Ling

Spine segmentation, based on ultrasound volume projection imaging (VPI), plays a vital role for intelligent scoliosis diagnosis in clinical applications. However, this task faces several significant challenges. Firstly, the global contextual knowledge of spines may not be well-learned if we neglect the high spatial correlation of different bone features. Secondly, the spine bones contain rich structural knowledge regarding their shapes and positions, which deserves to be encoded into the segmentation process. To address these challenges, we propose a novel scale-adaptive structure-aware network (SA$^{2}$Net) for effective spine segmentation. First, we propose a scale-adaptive complementary strategy to learn the cross-dimensional long-distance correlation features for spinal images. Second, motivated by the consistency between multi-head self-attention in Transformers and semantic level affinity, we propose structure-affinity transformation to transform semantic features with class-specific affinity and combine it with a Transformer decoder for structure-aware reasoning. In addition, we adopt a feature mixing loss aggregation method to enhance model training. This method improves the robustness and accuracy of the segmentation process. The experimental results demonstrate that our SA$^{2}$Net achieves superior segmentation performance compared to other state-of-the-art methods. Moreover, the adaptability of SA$^{2}$Net to various backbones enhances its potential as a promising tool for advanced scoliosis diagnosis using intelligent spinal image analysis. The code and experimental demo are available at https://github.com/taetiseo09/SA2Net.

comment: Accepted by Computerized Medical Imaging and Graphics (CMIG)

☆ Analysis of the Robustness of an Edge Detector Based on Cellular Automata Optimized by Particle Swarm

Vinícius Ferraria, Eurico Ruivo

The edge detection task is essential in image processing aiming to extract relevant information from an image. One recurring problem in this task is the weaknesses found in some detectors, such as the difficulty in detecting loose edges and the lack of context to extract relevant information from specific problems. To address these weaknesses and adapt the detector to the properties of an image, an adaptable detector described by two-dimensional cellular automaton and optimized by meta-heuristic combined with transfer learning techniques was developed. This study aims to analyze the impact of expanding the search space of the optimization phase and the robustness of the adaptability of the detector in identifying edges of a set of natural images and specialized subsets extracted from the same image set. The results obtained prove that expanding the search space of the optimization phase was not effective for the chosen image set. The study also analyzed the adaptability of the model through a series of experiments and validation techniques and found that, regardless of the validation, the model was able to adapt to the input and the transfer learning techniques applied to the model showed no significant improvements.

☆ Counteracting Matthew Effect in Self-Improvement of LVLMs through Head-Tail Re-balancing

Xin Guo, Zhiheng Xi, Yiwen Ding, Yitao Zhai, Xiaowei Shi, Xunliang Cai, Tao Gui, Qi Zhang, Xuanjing Huang

Self-improvement has emerged as a mainstream paradigm for advancing the reasoning capabilities of large vision-language models (LVLMs), where models explore and learn from successful trajectories iteratively. However, we identify a critical issue during this process: the model excels at generating high-quality trajectories for simple queries (i.e., head data) but struggles with more complex ones (i.e., tail data). This leads to an imbalanced optimization that drives the model to prioritize simple reasoning skills, while hindering its ability to tackle more complex reasoning tasks. Over iterations, this imbalance becomes increasingly pronounced--a dynamic we term the "Matthew effect"--which ultimately hinders further model improvement and leads to performance bottlenecks. To counteract this challenge, we introduce four efficient strategies from two perspectives: distribution-reshaping and trajectory-resampling, to achieve head-tail re-balancing during the exploration-and-learning self-improvement process. Extensive experiments on Qwen2-VL-7B-Instruct and InternVL2.5-4B models across visual reasoning tasks demonstrate that our methods consistently improve visual reasoning capabilities, outperforming vanilla self-improvement by 3.86 points on average.

comment: Preprint

☆ Representation-Level Counterfactual Calibration for Debiased Zero-Shot Recognition

Pei Peng, MingKun Xie, Hang Hao, Tong Jin, ShengJun Huang

Object-context shortcuts remain a persistent challenge in vision-language models, undermining zero-shot reliability when test-time scenes differ from familiar training co-occurrences. We recast this issue as a causal inference problem and ask: Would the prediction remain if the object appeared in a different environment? To answer this at inference time, we estimate object and background expectations within CLIP's representation space, and synthesize counterfactual embeddings by recombining object features with diverse alternative contexts sampled from external datasets, batch neighbors, or text-derived descriptions. By estimating the Total Direct Effect and simulating intervention, we further subtract background-only activation, preserving beneficial object-context interactions while mitigating hallucinated scores. Without retraining or prompt design, our method substantially improves both worst-group and average accuracy on context-sensitive benchmarks, establishing a new zero-shot state of the art. Beyond performance, our framework provides a lightweight representation-level counterfactual approach, offering a practical causal avenue for debiased and reliable multimodal reasoning.

☆ Towards Fine-Grained Vision-Language Alignment for Few-Shot Anomaly Detection

Yuanting Fan, Jun Liu, Xiaochen Chen, Bin-Bin Gao, Jian Li, Yong Liu, Jinlong Peng, Chengjie Wang

Few-shot anomaly detection (FSAD) methods identify anomalous regions with few known normal samples. Most existing methods rely on the generalization ability of pre-trained vision-language models (VLMs) to recognize potentially anomalous regions through feature similarity between text descriptions and images. However, due to the lack of detailed textual descriptions, these methods can only pre-define image-level descriptions to match each visual patch token to identify potential anomalous regions, which leads to the semantic misalignment between image descriptions and patch-level visual anomalies, achieving sub-optimal localization performance. To address the above issues, we propose the Multi-Level Fine-Grained Semantic Caption (MFSC) to provide multi-level and fine-grained textual descriptions for existing anomaly detection datasets with automatic construction pipeline. Based on the MFSC, we propose a novel framework named FineGrainedAD to improve anomaly localization performance, which consists of two components: Multi-Level Learnable Prompt (MLLP) and Multi-Level Semantic Alignment (MLSA). MLLP introduces fine-grained semantics into multi-level learnable prompts through automatic replacement and concatenation mechanism, while MLSA designs region aggregation strategy and multi-level alignment training to facilitate learnable prompts better align with corresponding visual regions. Experiments demonstrate that the proposed FineGrainedAD achieves superior overall performance in few-shot settings on MVTec-AD and VisA datasets.

comment: 12 pages, 7 figures

☆ PointSt3R: Point Tracking through 3D Grounded Correspondence

Rhodri Guerrier, Adam W. Harley, Dima Damen

Recent advances in foundational 3D reconstruction models, such as DUSt3R and MASt3R, have shown great potential in 2D and 3D correspondence in static scenes. In this paper, we propose to adapt them for the task of point tracking through 3D grounded correspondence. We first demonstrate that these models are competitive point trackers when focusing on static points, present in current point tracking benchmarks ($+33.5\%$ on EgoPoints vs. CoTracker2). We propose to combine the reconstruction loss with training for dynamic correspondence along with a visibility head, and fine-tuning MASt3R for point tracking using a relatively small amount of synthetic data. Importantly, we only train and evaluate on pairs of frames where one contains the query point, effectively removing any temporal context. Using a mix of dynamic and static point correspondences, we achieve competitive or superior point tracking results on four datasets (e.g. competitive on TAP-Vid-DAVIS 73.8 $\delta_{avg}$ / 85.8\% occlusion acc. for PointSt3R compared to 75.7 / 88.3\% for CoTracker2; and significantly outperform CoTracker3 on EgoPoints 61.3 vs 54.2 and RGB-S 87.0 vs 82.8). We also present results on 3D point tracking along with several ablations on training datasets and percentage of dynamic correspondences.

comment: http://rhodriguerrier.github.io/PointSt3R

☆ A-TPT: Angular Diversity Calibration Properties for Test-Time Prompt Tuning of Vision-Language Models

Shihab Aaqil Ahamed, Udaya S. K. P. Miriya Thanthrige, Ranga Rodrigo, Muhammad Haris Khan

Test-time prompt tuning (TPT) has emerged as a promising technique for adapting large vision-language models (VLMs) to unseen tasks without relying on labeled data. However, the lack of dispersion between textual features can hurt calibration performance, which raises concerns about VLMs' reliability, trustworthiness, and safety. Current TPT approaches primarily focus on improving prompt calibration by either maximizing average textual feature dispersion or enforcing orthogonality constraints to encourage angular separation. However, these methods may not always have optimal angular separation between class-wise textual features, which implies overlooking the critical role of angular diversity. To address this, we propose A-TPT, a novel TPT framework that introduces angular diversity to encourage uniformity in the distribution of normalized textual features induced by corresponding learnable prompts. This uniformity is achieved by maximizing the minimum pairwise angular distance between features on the unit hypersphere. We show that our approach consistently surpasses state-of-the-art TPT methods in reducing the aggregate average calibration error while maintaining comparable accuracy through extensive experiments with various backbones on different datasets. Notably, our approach exhibits superior zero-shot calibration performance on natural distribution shifts and generalizes well to medical datasets. We provide extensive analyses, including theoretical aspects, to establish the grounding of A-TPT. These results highlight the potency of promoting angular diversity to achieve well-dispersed textual features, significantly improving VLM calibration during test-time adaptation. Our code will be made publicly available.

comment: 23 pages, 14 figures

☆ LoCoT2V-Bench: A Benchmark for Long-Form and Complex Text-to-Video Generation

Xiangqing Zheng, Chengyue Wu, Kehai Chen, Min Zhang

Recently text-to-video generation has made impressive progress in producing short, high-quality clips, but evaluating long-form outputs remains a major challenge especially when processing complex prompts. Existing benchmarks mostly rely on simplified prompts and focus on low-level metrics, overlooking fine-grained alignment with prompts and abstract dimensions such as narrative coherence and thematic expression. To address these gaps, we propose LoCoT2V-Bench, a benchmark specifically designed for long video generation (LVG) under complex input conditions. Based on various real-world videos, LoCoT2V-Bench introduces a suite of realistic and complex prompts incorporating elements like scene transitions and event dynamics. Moreover, it constructs a multi-dimensional evaluation framework that includes our newly proposed metrics such as event-level alignment, fine-grained temporal consistency, content clarity, and the Human Expectation Realization Degree (HERD) that focuses on more abstract attributes like narrative flow, emotional response, and character development. Using this framework, we conduct a comprehensive evaluation of nine representative LVG models, finding that while current methods perform well on basic visual and temporal aspects, they struggle with inter-event consistency, fine-grained alignment, and high-level thematic adherence, etc. Overall, LoCoT2V-Bench provides a comprehensive and reliable platform for evaluating long-form complex text-to-video generation and highlights critical directions for future method improvement.

☆ EEG-Driven Image Reconstruction with Saliency-Guided Diffusion Models

Igor Abramov, Ilya Makarov

Existing EEG-driven image reconstruction methods often overlook spatial attention mechanisms, limiting fidelity and semantic coherence. To address this, we propose a dual-conditioning framework that combines EEG embeddings with spatial saliency maps to enhance image generation. Our approach leverages the Adaptive Thinking Mapper (ATM) for EEG feature extraction and fine-tunes Stable Diffusion 2.1 via Low-Rank Adaptation (LoRA) to align neural signals with visual semantics, while a ControlNet branch conditions generation on saliency maps for spatial control. Evaluated on THINGS-EEG, our method achieves a significant improvement in the quality of low- and high-level image features over existing approaches. Simultaneously, strongly aligning with human visual attention. The results demonstrate that attentional priors resolve EEG ambiguities, enabling high-fidelity reconstructions with applications in medical diagnostics and neuroadaptive interfaces, advancing neural decoding through efficient adaptation of pre-trained diffusion models.

comment: Demo paper

☆ SPG-CDENet: Spatial Prior-Guided Cross Dual Encoder Network for Multi-Organ Segmentation

Xizhi Tian, Changjun Zhou, Yulin. Yang

Multi-organ segmentation is a critical task in computer-aided diagnosis. While recent deep learning methods have achieved remarkable success in image segmentation, huge variations in organ size and shape challenge their effectiveness in multi-organ segmentation. To address these challenges, we propose a Spatial Prior-Guided Cross Dual Encoder Network (SPG-CDENet), a novel two-stage segmentation paradigm designed to improve multi-organ segmentation accuracy. Our SPG-CDENet consists of two key components: a spatial prior network and a cross dual encoder network. The prior network generates coarse localization maps that delineate the approximate ROI, serving as spatial guidance for the dual encoder network. The cross dual encoder network comprises four essential components: a global encoder, a local encoder, a symmetric cross-attention module, and a flow-based decoder. The global encoder captures global semantic features from the entire image, while the local encoder focuses on features from the prior network. To enhance the interaction between the global and local encoders, a symmetric cross-attention module is proposed across all layers of the encoders to fuse and refine features. Furthermore, the flow-based decoder directly propagates high-level semantic features from the final encoder layer to all decoder layers, maximizing feature preservation and utilization. Extensive qualitative and quantitative experiments on two public datasets demonstrate the superior performance of SPG-CDENet compared to existing segmentation methods. Furthermore, ablation studies further validate the effectiveness of the proposed modules in improving segmentation accuracy.

☆ CorVS: Person Identification via Video Trajectory-Sensor Correspondence in a Real-World Warehouse

Kazuma Kano, Yuki Mori, Shin Katayama, Kenta Urano, Takuro Yonezawa, Nobuo Kawaguchi

Worker location data is key to higher productivity in industrial sites. Cameras are a promising tool for localization in logistics warehouses since they also offer valuable environmental contexts such as package status. However, identifying individuals with only visual data is often impractical. Accordingly, several prior studies identified people in videos by comparing their trajectories and wearable sensor measurements. While this approach has advantages such as independence from appearance, the existing methods may break down under real-world conditions. To overcome this challenge, we propose CorVS, a novel data-driven person identification method based on correspondence between visual tracking trajectories and sensor measurements. Firstly, our deep learning model predicts correspondence probabilities and reliabilities for every pair of a trajectory and sensor measurements. Secondly, our algorithm matches the trajectories and sensor measurements over time using the predicted probabilities and reliabilities. We developed a dataset with actual warehouse operations and demonstrated the method's effectiveness for real-world applications.

comment: 7 pages, 3 figures, accepted to IPIN 2025

☆ AgriGS-SLAM: Orchard Mapping Across Seasons via Multi-View Gaussian Splatting SLAM

Mirko Usuelli, David Rapado-Rincon, Gert Kootstra, Matteo Matteucci

Autonomous robots in orchards require real-time 3D scene understanding despite repetitive row geometry, seasonal appearance changes, and wind-driven foliage motion. We present AgriGS-SLAM, a Visual--LiDAR SLAM framework that couples direct LiDAR odometry and loop closures with multi-camera 3D Gaussian Splatting (3DGS) rendering. Batch rasterization across complementary viewpoints recovers orchard structure under occlusions, while a unified gradient-driven map lifecycle executed between keyframes preserves fine details and bounds memory. Pose refinement is guided by a probabilistic LiDAR-based depth consistency term, back-propagated through the camera projection to tighten geometry-appearance coupling. We deploy the system on a field platform in apple and pear orchards across dormancy, flowering, and harvesting, using a standardized trajectory protocol that evaluates both training-view and novel-view synthesis to reduce 3DGS overfitting in evaluation. Across seasons and sites, AgriGS-SLAM delivers sharper, more stable reconstructions and steadier trajectories than recent state-of-the-art 3DGS-SLAM baselines while maintaining real-time performance on-tractor. While demonstrated in orchard monitoring, the approach can be applied to other outdoor domains requiring robust multimodal perception.

☆ GLYPH-SR: Can We Achieve Both High-Quality Image Super-Resolution and High-Fidelity Text Recovery via VLM-guided Latent Diffusion Model? ICLR 2026

Mingyu Sung, Seungjae Ham, Kangwoo Kim, Yeokyoung Yoon, Sangseok Yun, Il-Min Kim, Jae-Mo Kang

Image super-resolution(SR) is fundamental to many vision system-from surveillance and autonomy to document analysis and retail analytics-because recovering high-frequency details, especially scene-text, enables reliable downstream perception. Scene-text, i.e., text embedded in natural images such as signs, product labels, and storefronts, often carries the most actionable information; when characters are blurred or hallucinated, optical character recognition(OCR) and subsequent decisions fail even if the rest of the image appears sharp. Yet previous SR research has often been tuned to distortion (PSNR/SSIM) or learned perceptual metrics (LIPIS, MANIQA, CLIP-IQA, MUSIQ) that are largely insensitive to character-level errors. Furthermore, studies that do address text SR often focus on simplified benchmarks with isolated characters, overlooking the challenges of text within complex natural scenes. As a result, scene-text is effectively treated as generic texture. For SR to be effective in practical deployments, it is therefore essential to explicitly optimize for both text legibility and perceptual quality. We present GLYPH-SR, a vision-language-guided diffusion framework that aims to achieve both objectives jointly. GLYPH-SR utilizes a Text-SR Fusion ControlNet(TS-ControlNet) guided by OCR data, and a ping-pong scheduler that alternates between text- and scene-centric guidance. To enable targeted text restoration, we train these components on a synthetic corpus while keeping the main SR branch frozen. Across SVT, SCUT-CTW1500, and CUTE80 at x4, and x8, GLYPH-SR improves OCR F1 by up to +15.18 percentage points over diffusion/GAN baseline (SVT x8, OpenOCR) while maintaining competitive MANIQA, CLIP-IQA, and MUSIQ. GLYPH-SR is designed to satisfy both objectives simultaneously-high readability and high visual realism-delivering SR that looks right and reds right.

comment: 11 pages, 6 figures. Includes supplementary material. Under review as a conference paper at ICLR 2026

☆ A Hybrid Framework Bridging CNN and ViT based on Theory of Evidence for Diabetic Retinopathy Grading

Junlai Qiu, Yunzhu Chen, Hao Zheng, Yawen Huang, Yuexiang Li

Diabetic retinopathy (DR) is a leading cause of vision loss among middle-aged and elderly people, which significantly impacts their daily lives and mental health. To improve the efficiency of clinical screening and enable the early detection of DR, a variety of automated DR diagnosis systems have been recently established based on convolutional neural network (CNN) or vision Transformer (ViT). However, due to the own shortages of CNN / ViT, the performance of existing methods using single-type backbone has reached a bottleneck. One potential way for the further improvements is integrating different kinds of backbones, which can fully leverage the respective strengths of them (\emph{i.e.,} the local feature extraction capability of CNN and the global feature capturing ability of ViT). To this end, we propose a novel paradigm to effectively fuse the features extracted by different backbones based on the theory of evidence. Specifically, the proposed evidential fusion paradigm transforms the features from different backbones into supporting evidences via a set of deep evidential networks. With the supporting evidences, the aggregated opinion can be accordingly formed, which can be used to adaptively tune the fusion pattern between different backbones and accordingly boost the performance of our hybrid model. We evaluated our method on two publicly available DR grading datasets. The experimental results demonstrate that our hybrid model not only improves the accuracy of DR grading, compared to the state-of-the-art frameworks, but also provides the excellent interpretability for feature fusion and decision-making.

☆ Exploring the correlation between the type of music and the emotions evoked: A study using subjective questionnaires and EEG

Jelizaveta Jankowska, Bożena Kostek, Fernando Alonso-Fernandez, Prayag Tiwari

The subject of this work is to check how different types of music affect human emotions. While listening to music, a subjective survey and brain activity measurements were carried out using an EEG helmet. The aim is to demonstrate the impact of different music genres on emotions. The research involved a diverse group of participants of different gender and musical preferences. This had the effect of capturing a wide range of emotional responses to music. After the experiment, a relationship analysis of the respondents' questionnaires with EEG signals was performed. The analysis revealed connections between emotions and observed brain activity.

comment: Published at IWAIPR 2025 conference

☆ Towards Realistic Earth-Observation Constellation Scheduling: Benchmark and Methodology

Luting Wang, Yinghao Xiang, Hongliang Huang, Dongjun Li, Chen Gao, Si Liu

Agile Earth Observation Satellites (AEOSs) constellations offer unprecedented flexibility for monitoring the Earth's surface, but their scheduling remains challenging under large-scale scenarios, dynamic environments, and stringent constraints. Existing methods often simplify these complexities, limiting their real-world performance. We address this gap with a unified framework integrating a standardized benchmark suite and a novel scheduling model. Our benchmark suite, AEOS-Bench, contains $3,907$ finely tuned satellite assets and $16,410$ scenarios. Each scenario features $1$ to $50$ satellites and $50$ to $300$ imaging tasks. These scenarios are generated via a high-fidelity simulation platform, ensuring realistic satellite behavior such as orbital dynamics and resource constraints. Ground truth scheduling annotations are provided for each scenario. To our knowledge, AEOS-Bench is the first large-scale benchmark suite tailored for realistic constellation scheduling. Building upon this benchmark, we introduce AEOS-Former, a Transformer-based scheduling model that incorporates a constraint-aware attention mechanism. A dedicated internal constraint module explicitly models the physical and operational limits of each satellite. Through simulation-based iterative learning, AEOS-Former adapts to diverse scenarios, offering a robust solution for AEOS constellation scheduling. Experimental results demonstrate that AEOS-Former outperforms baseline models in task completion and energy efficiency, with ablation studies highlighting the contribution of each component. Code and data are provided in https://github.com/buaa-colalab/AEOSBench.

☆ Leveraging Large-Scale Face Datasets for Deep Periocular Recognition via Ocular Cropping

Fernando Alonso-Fernandez, Kevin Hernandez-Diaz, Jose Maria Buades Rubio, Josef Bigun

We focus on ocular biometrics, specifically the periocular region (the area around the eye), which offers high discrimination and minimal acquisition constraints. We evaluate three Convolutional Neural Network architectures of varying depth and complexity to assess their effectiveness for periocular recognition. The networks are trained on 1,907,572 ocular crops extracted from the large-scale VGGFace2 database. This significantly contrasts with existing works, which typically rely on small-scale periocular datasets for training having only a few thousand images. Experiments are conducted with ocular images from VGGFace2-Pose, a subset of VGGFace2 containing in-the-wild face images, and the UFPR-Periocular database, which consists of selfies captured via mobile devices with user guidance on the screen. Due to the uncontrolled conditions of VGGFace2, the Equal Error Rates (EERs) obtained with ocular crops range from 9-15%, noticeably higher than the 3-6% EERs achieved using full-face images. In contrast, UFPR-Periocular yields significantly better performance (EERs of 1-2%), thanks to higher image quality and more consistent acquisition protocols. To the best of our knowledge, these are the lowest reported EERs on the UFPR dataset to date.

comment: Published at IWAIPR 2025 conference

☆ Beyond Imitation: Constraint-Aware Trajectory Generation with Flow Matching For End-to-End Autonomous Driving

Lin Liu, Guanyi Yu, Ziying Song, Junqiao Li, Caiyan Jia, Feiyang Jia, Peiliang Wu, Yandan Luo

Planning is a critical component of end-to-end autonomous driving. However, prevailing imitation learning methods often suffer from mode collapse, failing to produce diverse trajectory hypotheses. Meanwhile, existing generative approaches struggle to incorporate crucial safety and physical constraints directly into the generative process, necessitating an additional optimization stage to refine their outputs. To address these limitations, we propose CATG, a novel planning framework that leverages Constrained Flow Matching. Concretely, CATG explicitly models the flow matching process, which inherently mitigates mode collapse and allows for flexible guidance from various conditioning signals. Our primary contribution is the novel imposition of explicit constraints directly within the flow matching process, ensuring that the generated trajectories adhere to vital safety and kinematic rules. Secondly, CATG parameterizes driving aggressiveness as a control signal during generation, enabling precise manipulation of trajectory style. Notably, on the NavSim v2 challenge, CATG achieved 2nd place with an EPDMS score of 51.31 and was honored with the Innovation Award.

☆ Exploring Complementarity and Explainability in CNNs for Periocular Verification Across Acquisition Distances

Fernando Alonso-Fernandez, Kevin Hernandez Diaz, Jose M. Buades, Kiran Raja, Josef Bigun

We study the complementarity of different CNNs for periocular verification at different distances on the UBIPr database. We train three architectures of increasing complexity (SqueezeNet, MobileNetv2, and ResNet50) on a large set of eye crops from VGGFace2. We analyse performance with cosine and chi2 metrics, compare different network initialisations, and apply score-level fusion via logistic regression. In addition, we use LIME heatmaps and Jensen-Shannon divergence to compare attention patterns of the CNNs. While ResNet50 consistently performs best individually, the fusion provides substantial gains, especially when combining all three networks. Heatmaps show that networks usually focus on distinct regions of a given image, which explains their complementarity. Our method significantly outperforms previous works on UBIPr, achieving a new state-of-the-art.

comment: Accepted at BIOSIG 2025 conference

☆ Revisiting Generative Infrared and Visible Image Fusion Based on Human Cognitive Laws NeurIPS 2025

Lin Guo, Xiaoqing Luo, Wei Xie, Zhancheng Zhang, Hui Li, Rui Wang, Zhenhua Feng, Xiaoning Song

Existing infrared and visible image fusion methods often face the dilemma of balancing modal information. Generative fusion methods reconstruct fused images by learning from data distributions, but their generative capabilities remain limited. Moreover, the lack of interpretability in modal information selection further affects the reliability and consistency of fusion results in complex scenarios. This manuscript revisits the essence of generative image fusion under the inspiration of human cognitive laws and proposes a novel infrared and visible image fusion method, termed HCLFuse. First, HCLFuse investigates the quantification theory of information mapping in unsupervised fusion networks, which leads to the design of a multi-scale mask-regulated variational bottleneck encoder. This encoder applies posterior probability modeling and information decomposition to extract accurate and concise low-level modal information, thereby supporting the generation of high-fidelity structural details. Furthermore, the probabilistic generative capability of the diffusion model is integrated with physical laws, forming a time-varying physical guidance mechanism that adaptively regulates the generation process at different stages, thereby enhancing the ability of the model to perceive the intrinsic structure of data and reducing dependence on data quality. Experimental results show that the proposed method achieves state-of-the-art fusion performance in qualitative and quantitative evaluations across multiple datasets and significantly improves semantic segmentation metrics. This fully demonstrates the advantages of this generative image fusion method, drawing inspiration from human cognition, in enhancing structural consistency and detail quality.

comment: NeurIPS 2025 spotlight

☆ Which Way Does Time Flow? A Psychophysics-Grounded Evaluation for Vision-Language Models

Shiho Matta, Lis Kanashiro Pereira, Peitao Han, Fei Cheng, Shigeru Kitazawa

Modern vision-language models (VLMs) excel at many multimodal tasks, yet their grasp of temporal information in video remains weak and, crucially, under-evaluated. We probe this gap with a deceptively simple but revealing challenge: judging the arrow of time (AoT)-whether a short clip is played forward or backward. We introduce AoT-PsyPhyBENCH, a psychophysically validated benchmark that tests whether VLMs can infer temporal direction in natural videos using the same stimuli and behavioral baselines established for humans. Our comprehensive evaluation of open-weight and proprietary, reasoning and non-reasoning VLMs reveals that most models perform near chance, and even the best lag far behind human accuracy on physically irreversible processes (e.g., free fall, diffusion/explosion) and causal manual actions (division/addition) that humans recognize almost instantly. These results highlight a fundamental gap in current multimodal systems: while they capture rich visual-semantic correlations, they lack the inductive biases required for temporal continuity and causal understanding. We release the code and data for AoT-PsyPhyBENCH to encourage further progress in the physical and temporal reasoning capabilities of VLMs.

comment: 10 pages

☆ OmniLayout: Enabling Coarse-to-Fine Learning with LLMs for Universal Document Layout Generation

Hengrui Kang, Zhuangcheng Gu, Zhiyuan Zhao, Zichen Wen, Bin Wang, Weijia Li, Conghui He

Document AI has advanced rapidly and is attracting increasing attention. Yet, while most efforts have focused on document layout analysis (DLA), its generative counterpart, document layout generation, remains underexplored. A major obstacle lies in the scarcity of diverse layouts: academic papers with Manhattan-style structures dominate existing studies, while open-world genres such as newspapers and magazines remain severely underrepresented. To address this gap, we curate OmniLayout-1M, the first million-scale dataset of diverse document layouts, covering six common document types and comprising contemporary layouts collected from multiple sources. Moreover, since existing methods struggle in complex domains and often fail to arrange long sequences coherently, we introduce OmniLayout-LLM, a 0.5B model with designed two-stage Coarse-to-Fine learning paradigm: 1) learning universal layout principles from OmniLayout-1M with coarse category definitions, and 2) transferring the knowledge to a specific domain with fine-grained annotations. Extensive experiments demonstrate that our approach achieves strong performance on multiple domains in M$^{6}$Doc dataset, substantially surpassing both existing layout generation experts and several latest general-purpose LLMs. Our code, models, and dataset will be publicly released.

comment: TL;DR: With OmniLayout-1M dataset and LLM-based coarse-to-fine learning, we enable universal and diverse document layout generation

☆ Developing a Multi-task Ensemble Geometric Deep Network for Supply Chain Sustainability and Risk Management

Mehdi Khaleghi, Nastaran Khaleghi, Sobhan Sheykhivand, Sebelan Danishvar

The sustainability of supply chain plays a key role in achieving optimal performance in controlling the supply chain. The management of risks that occur in a supply chain is a fundamental problem for the purpose of developing the sustainability of the network and elevating the performance efficiency of the supply chain. The correct classification of products is another essential element in a sustainable supply chain. Acknowledging recent breakthroughs in the context of deep networks, several architectural options have been deployed to analyze supply chain datasets. A novel geometric deep network is used to propose an ensemble deep network. The proposed Chebyshev ensemble geometric network (Ch-EGN) is a hybrid convolutional and geometric deep learning. This network is proposed to leverage the information dependencies in supply chain to derive invisible states of samples in the database. The functionality of the proposed deep network is assessed on the two different databases. The SupplyGraph Dataset and DataCo are considered in this research. The prediction of delivery status of DataCo supply chain is done for risk administration. The product classification and edge classification are performed using the SupplyGraph database to enhance the sustainability of the supply network. An average accuracy of 98.95% is obtained for the ensemble network for risk management. The average accuracy of 100% and 98.07% are obtained for sustainable supply chain in terms of 5 product group classification and 4 product relation classification, respectively. The average accuracy of 92.37% is attained for 25 company relation classification. The results confirm an average improvement and efficiency of the proposed method compared to the state-of-the-art approaches.

☆ Sketch2PoseNet: Efficient and Generalized Sketch to 3D Human Pose Prediction SIGGRAPH

Li Wang, Yiyu Zhuang, Yanwen Wang, Xun Cao, Chuan Guo, Xinxin Zuo, Hao Zhu

3D human pose estimation from sketches has broad applications in computer animation and film production. Unlike traditional human pose estimation, this task presents unique challenges due to the abstract and disproportionate nature of sketches. Previous sketch-to-pose methods, constrained by the lack of large-scale sketch-3D pose annotations, primarily relied on optimization with heuristic rules-an approach that is both time-consuming and limited in generalizability. To address these challenges, we propose a novel approach leveraging a "learn from synthesis" strategy. First, a diffusion model is trained to synthesize sketch images from 2D poses projected from 3D human poses, mimicking disproportionate human structures in sketches. This process enables the creation of a synthetic dataset, SKEP-120K, consisting of 120k accurate sketch-3D pose annotation pairs across various sketch styles. Building on this synthetic dataset, we introduce an end-to-end data-driven framework for estimating human poses and shapes from diverse sketch styles. Our framework combines existing 2D pose detectors and generative diffusion priors for sketch feature extraction with a feed-forward neural network for efficient 2D pose estimation. Multiple heuristic loss functions are incorporated to guarantee geometric coherence between the derived 3D poses and the detected 2D poses while preserving accurate self-contacts. Qualitative, quantitative, and subjective evaluations collectively show that our model substantially surpasses previous ones in both estimation accuracy and speed for sketch-to-pose tasks.

comment: SIGGRAPH Asia 2025

☆ ConceptScope: Characterizing Dataset Bias via Disentangled Visual Concepts NeurIPS 2025

Jinho Choi, Hyesu Lim, Steffen Schneider, Jaegul Choo

Dataset bias, where data points are skewed to certain concepts, is ubiquitous in machine learning datasets. Yet, systematically identifying these biases is challenging without costly, fine-grained attribute annotations. We present ConceptScope, a scalable and automated framework for analyzing visual datasets by discovering and quantifying human-interpretable concepts using Sparse Autoencoders trained on representations from vision foundation models. ConceptScope categorizes concepts into target, context, and bias types based on their semantic relevance and statistical correlation to class labels, enabling class-level dataset characterization, bias identification, and robustness evaluation through concept-based subgrouping. We validate that ConceptScope captures a wide range of visual concepts, including objects, textures, backgrounds, facial attributes, emotions, and actions, through comparisons with annotated datasets. Furthermore, we show that concept activations produce spatial attributions that align with semantically meaningful image regions. ConceptScope reliably detects known biases (e.g., background bias in Waterbirds) and uncovers previously unannotated ones (e.g, co-occurring objects in ImageNet), offering a practical tool for dataset auditing and model diagnostics.

comment: Published in the Thirty-Ninth Conference on Neural Information Processing Systems (NeurIPS 2025)

☆ MoTDiff: High-resolution Motion Trajectory estimation from a single blurred image using Diffusion models

Wontae Choi, Jaelin Lee, Hyung Sup Yun, Byeungwoo Jeon, Il Yong Chun

Accurate estimation of motion information is crucial in diverse computational imaging and computer vision applications. Researchers have investigated various methods to extract motion information from a single blurred image, including blur kernels and optical flow. However, existing motion representations are often of low quality, i.e., coarse-grained and inaccurate. In this paper, we propose the first high-resolution (HR) Motion Trajectory estimation framework using Diffusion models (MoTDiff). Different from existing motion representations, we aim to estimate an HR motion trajectory with high-quality from a single motion-blurred image. The proposed MoTDiff consists of two key components: 1) a new conditional diffusion framework that uses multi-scale feature maps extracted from a single blurred image as a condition, and 2) a new training method that can promote precise identification of a fine-grained motion trajectory, consistent estimation of overall shape and position of a motion path, and pixel connectivity along a motion trajectory. Our experiments demonstrate that the proposed MoTDiff can outperform state-of-the-art methods in both blind image deblurring and coded exposure photography applications.

comment: 10 pages, 6 figures

☆ Self-localization on a 3D map by fusing global and local features from a monocular camera

Satoshi Kikuch, Masaya Kato, Tsuyoshi Tasaki

Self-localization on a 3D map by using an inexpensive monocular camera is required to realize autonomous driving. Self-localization based on a camera often uses a convolutional neural network (CNN) that can extract local features that are calculated by nearby pixels. However, when dynamic obstacles, such as people, are present, CNN does not work well. This study proposes a new method combining CNN with Vision Transformer, which excels at extracting global features that show the relationship of patches on whole image. Experimental results showed that, compared to the state-of-the-art method (SOTA), the accuracy improvement rate in a CG dataset with dynamic obstacles is 1.5 times higher than that without dynamic obstacles. Moreover, the self-localization error of our method is 20.1% smaller than that of SOTA on public datasets. Additionally, our robot using our method can localize itself with 7.51cm error on average, which is more accurate than SOTA.

☆ CRAG-MM: Multi-modal Multi-turn Comprehensive RAG Benchmark

Jiaqi Wang, Xiao Yang, Kai Sun, Parth Suresh, Sanat Sharma, Adam Czyzewski, Derek Andersen, Surya Appini, Arkav Banerjee, Sajal Choudhary, Shervin Ghasemlou, Ziqiang Guan, Akil Iyer, Haidar Khan, Lingkun Kong, Roy Luo, Tiffany Ma, Zhen Qiao, David Tran, Wenfang Xu, Skyler Yeatman, Chen Zhou, Gunveer Gujral, Yinglong Xia, Shane Moon, Nicolas Scheffer, Nirav Shah, Eun Chang, Yue Liu, Florian Metze, Tammy Stark, Zhaleh Feizollahi, Andrea Jessee, Mangesh Pujari, Ahmed Aly, Babak Damavandi, Rakesh Wanga, Anuj Kumar, Rohit Patel, Wen-tau Yih, Xin Luna Dong

Wearable devices such as smart glasses are transforming the way people interact with their surroundings, enabling users to seek information regarding entities in their view. Multi-Modal Retrieval-Augmented Generation (MM-RAG) plays a key role in supporting such questions, yet there is still no comprehensive benchmark for this task, especially regarding wearables scenarios. To fill this gap, we present CRAG-MM -- a Comprehensive RAG benchmark for Multi-modal Multi-turn conversations. CRAG-MM contains a diverse set of 6.5K (image, question, answer) triplets and 2K visual-based multi-turn conversations across 13 domains, including 6.2K egocentric images designed to mimic captures from wearable devices. We carefully constructed the questions to reflect real-world scenarios and challenges, including five types of image-quality issues, six question types, varying entity popularity, differing information dynamism, and different conversation turns. We design three tasks: single-source augmentation, multi-source augmentation, and multi-turn conversations -- each paired with an associated retrieval corpus and APIs for both image-KG retrieval and webpage retrieval. Our evaluation shows that straightforward RAG approaches achieve only 32% and 43% truthfulness on CRAG-MM single- and multi-turn QA, respectively, whereas state-of-the-art industry solutions have similar quality (32%/45%), underscoring ample room for improvement. The benchmark has hosted KDD Cup 2025, attracting about 1K participants and 5K submissions, with winning solutions improving baseline performance by 28%, highlighting its early impact on advancing the field.

☆ Detecting Unauthorized Vehicles using Deep Learning for Smart Cities: A Case Study on Bangladesh

Sudipto Das Sukanto, Diponker Roy, Fahim Shakil, Nirjhar Singha, Abdullah Asik, Aniket Joarder, Mridha Md Nafis Fuad, Muhammad Ibrahim

Modes of transportation vary across countries depending on geographical location and cultural context. In South Asian countries rickshaws are among the most common means of local transport. Based on their mode of operation, rickshaws in cities across Bangladesh can be broadly classified into non-auto (pedal-powered) and auto-rickshaws (motorized). Monitoring the movement of auto-rickshaws is necessary as traffic rules often restrict auto-rickshaws from accessing certain routes. However, existing surveillance systems make it quite difficult to monitor them due to their similarity to other vehicles, especially non-auto rickshaws whereas manual video analysis is too time-consuming. This paper presents a machine learning-based approach to automatically detect auto-rickshaws in traffic images. In this system, we used real-time object detection using the YOLOv8 model. For training purposes, we prepared a set of 1,730 annotated images that were captured under various traffic conditions. The results show that our proposed model performs well in real-time auto-rickshaw detection and offers an mAP50 of 83.447% and binary precision and recall values above 78%, demonstrating its effectiveness in handling both dense and sparse traffic scenarios. The dataset has been publicly released for further research.

comment: 16 pages

☆ MV-MLM: Bridging Multi-View Mammography and Language for Breast Cancer Diagnosis and Risk Prediction ICCV 2025

Shunjie-Fabian Zheng, Hyeonjun Lee, Thijs Kooi, Ali Diba

Large annotated datasets are essential for training robust Computer-Aided Diagnosis (CAD) models for breast cancer detection or risk prediction. However, acquiring such datasets with fine-detailed annotation is both costly and time-consuming. Vision-Language Models (VLMs), such as CLIP, which are pre-trained on large image-text pairs, offer a promising solution by enhancing robustness and data efficiency in medical imaging tasks. This paper introduces a novel Multi-View Mammography and Language Model for breast cancer classification and risk prediction, trained on a dataset of paired mammogram images and synthetic radiology reports. Our MV-MLM leverages multi-view supervision to learn rich representations from extensive radiology data by employing cross-modal self-supervision across image-text pairs. This includes multiple views and the corresponding pseudo-radiology reports. We propose a novel joint visual-textual learning strategy to enhance generalization and accuracy performance over different data types and tasks to distinguish breast tissues or cancer characteristics(calcification, mass) and utilize these patterns to understand mammography images and predict cancer risk. We evaluated our method on both private and publicly available datasets, demonstrating that the proposed model achieves state-of-the-art performance in three classification tasks: (1) malignancy classification, (2) subtype classification, and (3) image-based cancer risk prediction. Furthermore, the model exhibits strong data efficiency, outperforming existing fully supervised or VLM baselines while trained on synthetic text reports and without the need for actual radiology reports.

comment: Accepted to Computer Vision for Automated Medical Diagnosis (CVAMD) Workshop at ICCV 2025

☆ BasicAVSR: Arbitrary-Scale Video Super-Resolution via Image Priors and Enhanced Motion Compensation

Wei Shang, Wanying Zhang, Shuhang Gu, Pengfei Zhu, Qinghua Hu, Dongwei Ren

Arbitrary-scale video super-resolution (AVSR) aims to enhance the resolution of video frames, potentially at various scaling factors, which presents several challenges regarding spatial detail reproduction, temporal consistency, and computational complexity. In this paper, we propose a strong baseline BasicAVSR for AVSR by integrating four key components: 1) adaptive multi-scale frequency priors generated from image Laplacian pyramids, 2) a flow-guided propagation unit to aggregate spatiotemporal information from adjacent frames, 3) a second-order motion compensation unit for more accurate spatial alignment of adjacent frames, and 4) a hyper-upsampling unit to generate scale-aware and content-independent upsampling kernels. To meet diverse application demands, we instantiate three propagation variants: (i) a unidirectional RNN unit for strictly online inference, (ii) a unidirectional RNN unit empowered with a limited lookahead that tolerates a small output delay, and (iii) a bidirectional RNN unit designed for offline tasks where computational resources are less constrained. Experimental results demonstrate the effectiveness and adaptability of our model across these different scenarios. Through extensive experiments, we show that BasicAVSR significantly outperforms existing methods in terms of super-resolution quality, generalization ability, and inference speed. Our work not only advances the state-of-the-art in AVSR but also extends its core components to multiple frameworks for diverse scenarios. The code is available at https://github.com/shangwei5/BasicAVSR.

comment: 13 pages, 10 figures, 5 tables

☆ StructLayoutFormer:Conditional Structured Layout Generation via Structure Serialization and Disentanglement

Xin Hu, Pengfei Xu, Jin Zhou, Hongbo Fu, Hui Huang

Structured layouts are preferable in many 2D visual contents (\eg, GUIs, webpages) since the structural information allows convenient layout editing. Computational frameworks can help create structured layouts but require heavy labor input. Existing data-driven approaches are effective in automatically generating fixed layouts but fail to produce layout structures. We present StructLayoutFormer, a novel Transformer-based approach for conditional structured layout generation. We use a structure serialization scheme to represent structured layouts as sequences. To better control the structures of generated layouts, we disentangle the structural information from the element placements. Our approach is the first data-driven approach that achieves conditional structured layout generation and produces realistic layout structures explicitly. We compare our approach with existing data-driven layout generation approaches by including post-processing for structure extraction. Extensive experiments have shown that our approach exceeds these baselines in conditional structured layout generation. We also demonstrate that our approach is effective in extracting and transferring layout structures. The code is publicly available at %\href{https://github.com/Teagrus/StructLayoutFormer} {https://github.com/Teagrus/StructLayoutFormer}.

☆ FullPart: Generating each 3D Part at Full Resolution

Lihe Ding, Shaocong Dong, Yaokun Li, Chenjian Gao, Xiao Chen, Rui Han, Yihao Kuang, Hong Zhang, Bo Huang, Zhanpeng Huang, Zibin Wang, Dan Xu, Tianfan Xue

Part-based 3D generation holds great potential for various applications. Previous part generators that represent parts using implicit vector-set tokens often suffer from insufficient geometric details. Another line of work adopts an explicit voxel representation but shares a global voxel grid among all parts; this often causes small parts to occupy too few voxels, leading to degraded quality. In this paper, we propose FullPart, a novel framework that combines both implicit and explicit paradigms. It first derives the bounding box layout through an implicit box vector-set diffusion process, a task that implicit diffusion handles effectively since box tokens contain little geometric detail. Then, it generates detailed parts, each within its own fixed full-resolution voxel grid. Instead of sharing a global low-resolution space, each part in our method - even small ones - is generated at full resolution, enabling the synthesis of intricate details. We further introduce a center-point encoding strategy to address the misalignment issue when exchanging information between parts of different actual sizes, thereby maintaining global coherence. Moreover, to tackle the scarcity of reliable part data, we present PartVerse-XL, the largest human-annotated 3D part dataset to date with 40K objects and 320K parts. Extensive experiments demonstrate that FullPart achieves state-of-the-art results in 3D part generation. We will release all code, data, and model to benefit future research in 3D part generation.

comment: Project page: https://fullpart3d.github.io

☆ Exploring Object-Aware Attention Guided Frame Association for RGB-D SLAM

Ali Caglayan, Nevrez Imamoglu, Oguzhan Guclu, Ali Osman Serhatoglu, Ahmet Burak Can, Ryosuke Nakamura

Attention models have recently emerged as a powerful approach, demonstrating significant progress in various fields. Visualization techniques, such as class activation mapping, provide visual insights into the reasoning of convolutional neural networks (CNNs). Using network gradients, it is possible to identify regions where the network pays attention during image recognition tasks. Furthermore, these gradients can be combined with CNN features to localize more generalizable, task-specific attentive (salient) regions within scenes. However, explicit use of this gradient-based attention information integrated directly into CNN representations for semantic object understanding remains limited. Such integration is particularly beneficial for visual tasks like simultaneous localization and mapping (SLAM), where CNN representations enriched with spatially attentive object locations can enhance performance. In this work, we propose utilizing task-specific network attention for RGB-D indoor SLAM. Specifically, we integrate layer-wise attention information derived from network gradients with CNN feature representations to improve frame association performance. Experimental results indicate improved performance compared to baseline methods, particularly for large environments.

comment: double-column 5 pages, 3 figures

☆ WOD-E2E: Waymo Open Dataset for End-to-End Driving in Challenging Long-tail Scenarios

Runsheng Xu, Hubert Lin, Wonseok Jeon, Hao Feng, Yuliang Zou, Liting Sun, John Gorman, Kate Tolstaya, Sarah Tang, Brandyn White, Ben Sapp, Mingxing Tan, Jyh-Jing Hwang, Drago Anguelov

Vision-based end-to-end (E2E) driving has garnered significant interest in the research community due to its scalability and synergy with multimodal large language models (MLLMs). However, current E2E driving benchmarks primarily feature nominal scenarios, failing to adequately test the true potential of these systems. Furthermore, existing open-loop evaluation metrics often fall short in capturing the multi-modal nature of driving or effectively evaluating performance in long-tail scenarios. To address these gaps, we introduce the Waymo Open Dataset for End-to-End Driving (WOD-E2E). WOD-E2E contains 4,021 driving segments (approximately 12 hours), specifically curated for challenging long-tail scenarios that that are rare in daily life with an occurring frequency of less than 0.03%. Concretely, each segment in WOD-E2E includes the high-level routing information, ego states, and 360-degree camera views from 8 surrounding cameras. To evaluate the E2E driving performance on these long-tail situations, we propose a novel open-loop evaluation metric: Rater Feedback Score (RFS). Unlike conventional metrics that measure the distance between predicted way points and the logs, RFS measures how closely the predicted trajectory matches rater-annotated trajectory preference labels. We have released rater preference labels for all WOD-E2E validation set segments, while the held out test set labels have been used for the 2025 WOD-E2E Challenge. Through our work, we aim to foster state of the art research into generalizable, robust, and safe end-to-end autonomous driving agents capable of handling complex real-world situations.

☆ JOGS: Joint Optimization of Pose Estimation and 3D Gaussian Splatting

Yuxuan Li, Tao Wang, Xianben Yang

Traditional novel view synthesis methods heavily rely on external camera pose estimation tools such as COLMAP, which often introduce computational bottlenecks and propagate errors. To address these challenges, we propose a unified framework that jointly optimizes 3D Gaussian points and camera poses without requiring pre-calibrated inputs. Our approach iteratively refines 3D Gaussian parameters and updates camera poses through a novel co-optimization strategy, ensuring simultaneous improvements in scene reconstruction fidelity and pose accuracy. The key innovation lies in decoupling the joint optimization into two interleaved phases: first, updating 3D Gaussian parameters via differentiable rendering with fixed poses, and second, refining camera poses using a customized 3D optical flow algorithm that incorporates geometric and photometric constraints. This formulation progressively reduces projection errors, particularly in challenging scenarios with large viewpoint variations and sparse feature distributions, where traditional methods struggle. Extensive evaluations on multiple datasets demonstrate that our approach significantly outperforms existing COLMAP-free techniques in reconstruction quality, and also surpasses the standard COLMAP-based baseline in general.

☆ OracleAgent: A Multimodal Reasoning Agent for Oracle Bone Script Research

Caoshuo Li, Zengmao Ding, Xiaobin Hu, Bang Li, Donghao Luo, Xu Peng, Taisong Jin, Yongge Liu, Shengwei Han, Jing Yang, Xiaoping He, Feng Gao, AndyPian Wu, SevenShu, Chaoyang Wang, Chengjie Wang

As one of the earliest writing systems, Oracle Bone Script (OBS) preserves the cultural and intellectual heritage of ancient civilizations. However, current OBS research faces two major challenges: (1) the interpretation of OBS involves a complex workflow comprising multiple serial and parallel sub-tasks, and (2) the efficiency of OBS information organization and retrieval remains a critical bottleneck, as scholars often spend substantial effort searching for, compiling, and managing relevant resources. To address these challenges, we present OracleAgent, the first agent system designed for the structured management and retrieval of OBS-related information. OracleAgent seamlessly integrates multiple OBS analysis tools, empowered by large language models (LLMs), and can flexibly orchestrate these components. Additionally, we construct a comprehensive domain-specific multimodal knowledge base for OBS, which is built through a rigorous multi-year process of data collection, cleaning, and expert annotation. The knowledge base comprises over 1.4M single-character rubbing images and 80K interpretation texts. OracleAgent leverages this resource through its multimodal tools to assist experts in retrieval tasks of character, document, interpretation text, and rubbing image. Extensive experiments demonstrate that OracleAgent achieves superior performance across a range of multimodal reasoning and generation tasks, surpassing leading mainstream multimodal large language models (MLLMs) (e.g., GPT-4o). Furthermore, our case study illustrates that OracleAgent can effectively assist domain experts, significantly reducing the time cost of OBS research. These results highlight OracleAgent as a significant step toward the practical deployment of OBS-assisted research and automated interpretation systems.

☆ EgoExo-Con: Exploring View-Invariant Video Temporal Understanding

Minjoon Jung, Junbin Xiao, Junghyun Kim, Byoung-Tak Zhang, Angela Yao

Can Video-LLMs achieve consistent temporal understanding when videos capture the same event from different viewpoints? To study this, we introduce EgoExo-Con (Consistency), a benchmark of comprehensively synchronized egocentric and exocentric video pairs with human-refined queries in natural language. EgoExo-Con emphasizes two temporal understanding tasks: Temporal Verification and Temporal Grounding. It evaluates not only correctness but consistency across viewpoints. Our analysis reveals two critical limitations of existing Video-LLMs: (1) models often fail to maintain consistency, with results far worse than their single-view performances. (2) When naively finetuned with synchronized videos of both viewpoints, the models show improved consistency but often underperform those trained on a single view. For improvements, we propose View-GRPO, a novel reinforcement learning framework that effectively strengthens view-specific temporal reasoning while encouraging consistent comprehension across viewpoints. Our method demonstrates its superiority over naive SFT and GRPO, especially for improving cross-view consistency. All resources will be made publicly available.

comment: project page: \url{https://minjoong507.github.io/projects/EgoExo-Con/}

☆ Security Risk of Misalignment between Text and Image in Multi-modal Model

Xiaosen Wang, Zhijin Ge, Shaokang Wang

Despite the notable advancements and versatility of multi-modal diffusion models, such as text-to-image models, their susceptibility to adversarial inputs remains underexplored. Contrary to expectations, our investigations reveal that the alignment between textual and Image modalities in existing diffusion models is inadequate. This misalignment presents significant risks, especially in the generation of inappropriate or Not-Safe-For-Work (NSFW) content. To this end, we propose a novel attack called Prompt-Restricted Multi-modal Attack (PReMA) to manipulate the generated content by modifying the input image in conjunction with any specified prompt, without altering the prompt itself. PReMA is the first attack that manipulates model outputs by solely creating adversarial images, distinguishing itself from prior methods that primarily generate adversarial prompts to produce NSFW content. Consequently, PReMA poses a novel threat to the integrity of multi-modal diffusion models, particularly in image-editing applications that operate with fixed prompts. Comprehensive evaluations conducted on image inpainting and style transfer tasks across various models confirm the potent efficacy of PReMA.

☆ Dynamic VLM-Guided Negative Prompting for Diffusion Models NeurIPS 2025

Hoyeon Chang, Seungjin Kim, Yoonseok Choi

We propose a novel approach for dynamic negative prompting in diffusion models that leverages Vision-Language Models (VLMs) to adaptively generate negative prompts during the denoising process. Unlike traditional Negative Prompting methods that use fixed negative prompts, our method generates intermediate image predictions at specific denoising steps and queries a VLM to produce contextually appropriate negative prompts. We evaluate our approach on various benchmark datasets and demonstrate the trade-offs between negative guidance strength and text-image alignment.

comment: 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: The First Workshop on Generative and Protective AI for Content Creation

☆ FlexICL: A Flexible Visual In-context Learning Framework for Elbow and Wrist Ultrasound Segmentation

Yuyue Zhou, Jessica Knight, Shrimanti Ghosh, Banafshe Felfeliyan, Jacob L. Jaremko, Abhilash R. Hareendranathan

Elbow and wrist fractures are the most common fractures in pediatric populations. Automatic segmentation of musculoskeletal structures in ultrasound (US) can improve diagnostic accuracy and treatment planning. Fractures appear as cortical defects but require expert interpretation. Deep learning (DL) can provide real-time feedback and highlight key structures, helping lightly trained users perform exams more confidently. However, pixel-wise expert annotations for training remain time-consuming and costly. To address this challenge, we propose FlexICL, a novel and flexible in-context learning (ICL) framework for segmenting bony regions in US images. We apply it to an intra-video segmentation setting, where experts annotate only a small subset of frames, and the model segments unseen frames. We systematically investigate various image concatenation techniques and training strategies for visual ICL and introduce novel concatenation methods that significantly enhance model performance with limited labeled data. By integrating multiple augmentation strategies, FlexICL achieves robust segmentation performance across four wrist and elbow US datasets while requiring only 5% of the training images. It outperforms state-of-the-art visual ICL models like Painter, MAE-VQGAN, and conventional segmentation models like U-Net and TransUNet by 1-27% Dice coefficient on 1,252 US sweeps. These initial results highlight the potential of FlexICL as an efficient and scalable solution for US image segmentation well suited for medical imaging use cases where labeled data is scarce.

☆ Do Students Debias Like Teachers? On the Distillability of Bias Mitigation Methods

Jiali Cheng, Chirag Agarwal, Hadi Amiri

Knowledge distillation (KD) is an effective method for model compression and transferring knowledge between models. However, its effect on model's robustness against spurious correlations that degrade performance on out-of-distribution data remains underexplored. This study investigates the effect of knowledge distillation on the transferability of ``debiasing'' capabilities from teacher models to student models on natural language inference (NLI) and image classification tasks. Through extensive experiments, we illustrate several key findings: (i) overall the debiasing capability of a model is undermined post-KD; (ii) training a debiased model does not benefit from injecting teacher knowledge; (iii) although the overall robustness of a model may remain stable post-distillation, significant variations can occur across different types of biases; and (iv) we pin-point the internal attention pattern and circuit that causes the distinct behavior post-KD. Given the above findings, we propose three effective solutions to improve the distillability of debiasing methods: developing high quality data for augmentation, implementing iterative knowledge distillation, and initializing student models with weights obtained from teacher models. To the best of our knowledge, this is the first study on the effect of KD on debiasing and its interenal mechanism at scale. Our findings provide understandings on how KD works and how to design better debiasing methods.

♻ ☆ Smoothing Slot Attention Iterations and Recurrences

Rongzhen Zhao, Wenyan Yang, Juho Kannala, Joni Pajarinen

Slot Attention (SA) and its variants lie at the heart of mainstream Object-Centric Learning (OCL). Objects in an image can be aggregated into respective slot vectors, by \textit{iteratively} refining cold-start query vectors, typically three times, via SA on image features. For video, such aggregation is \textit{recurrently} shared across frames, with queries cold-started on the first frame while transitioned from the previous frame's slots on non-first frames. However, the cold-start queries lack sample-specific cues thus hinder precise aggregation on the image or video's first frame; Also, non-first frames' queries are already sample-specific thus require transforms different from the first frame's aggregation. We address these issues for the first time with our \textit{SmoothSA}: (1) To smooth SA iterations on the image or video's first frame, we \textit{preheat} the cold-start queries with rich information of input features, via a tiny module self-distilled inside OCL; (2) To smooth SA recurrences across all video frames, we \textit{differentiate} the homogeneous transforms on the first and non-first frames, by using full and single iterations respectively. Comprehensive experiments on object discovery, recognition and downstream benchmarks validate our method's effectiveness. Further analyses intuitively illuminate how our method smooths SA iterations and recurrences. Our source code, model checkpoints and training logs are available on https://github.com/Genera1Z/SmoothSA.

♻ ☆ Predicting Video Slot Attention Queries from Random Slot-Feature Pairs

Rongzhen Zhao, Jian Li, Juho Kannala, Joni Pajarinen

Unsupervised video Object-Centric Learning (OCL) is promising as it enables object-level scene representation and dynamics modeling as we humans do. Mainstream video OCL methods adopt a recurrent architecture: An aggregator aggregates current video frame into object features, termed slots, under some queries; A transitioner transits current slots to queries for the next frame. This is an effective architecture but all existing implementations both (\textit{i1}) neglect to incorporate next frame features, the most informative source for query prediction, and (\textit{i2}) fail to learn transition dynamics, the knowledge essential for query prediction. To address these issues, we propose Random Slot-Feature pair for learning Query prediction (RandSF.Q): (\textit{t1}) We design a new transitioner to incorporate both slots and features, which provides more information for query prediction; (\textit{t2}) We train the transitioner to predict queries from slot-feature pairs randomly sampled from available recurrences, which drives it to learn transition dynamics. Experiments on scene representation demonstrate that our method surpass existing video OCL methods significantly, e.g., up to 10 points on object discovery, setting new state-of-the-art. Such superiority also benefits downstream tasks like dynamics modeling. Our core source code, model checkpoints and training logs are available on https://github.com/Genera1Z/RandSF.Q.

♻ ☆ Locality in Image Diffusion Models Emerges from Data Statistics

Artem Lukoianov, Chenyang Yuan, Justin Solomon, Vincent Sitzmann

Recent work has shown that the generalization ability of image diffusion models arises from the locality properties of the trained neural network. In particular, when denoising a particular pixel, the model relies on a limited neighborhood of the input image around that pixel, which, according to the previous work, is tightly related to the ability of these models to produce novel images. Since locality is central to generalization, it is crucial to understand why diffusion models learn local behavior in the first place, as well as the factors that govern the properties of locality patterns. In this work, we present evidence that the locality in deep diffusion models emerges as a statistical property of the image dataset and is not due to the inductive bias of convolutional neural networks, as suggested in previous work. Specifically, we demonstrate that an optimal parametric linear denoiser exhibits similar locality properties to deep neural denoisers. We show, both theoretically and experimentally, that this locality arises directly from pixel correlations present in the image datasets. Moreover, locality patterns are drastically different on specialized datasets, approximating principal components of the data's covariance. We use these insights to craft an analytical denoiser that better matches scores predicted by a deep diffusion model than prior expert-crafted alternatives. Our key takeaway is that while neural network architectures influence generation quality, their primary role is to capture locality patterns inherent in the data.

comment: 31 pages, 20 figures, 7 tables

♻ ☆ ScoreAdv: Score-based Targeted Generation of Natural Adversarial Examples via Diffusion Models

Chihan Huang, Hao Tang

Despite the success of deep learning across various domains, it remains vulnerable to adversarial attacks. Although many existing adversarial attack methods achieve high success rates, they typically rely on $\ell_{p}$-norm perturbation constraints, which do not align with human perceptual capabilities. Consequently, researchers have shifted their focus toward generating natural, unrestricted adversarial examples (UAEs). GAN-based approaches suffer from inherent limitations, such as poor image quality due to instability and mode collapse. Meanwhile, diffusion models have been employed for UAE generation, but they still rely on iterative PGD perturbation injection, without fully leveraging their central denoising capabilities. In this paper, we introduce a novel approach for generating UAEs based on diffusion models, named ScoreAdv. This method incorporates an interpretable adversarial guidance mechanism to gradually shift the sampling distribution towards the adversarial distribution, while using an interpretable saliency map to inject the visual information of a reference image into the generated samples. Notably, our method is capable of generating an unlimited number of natural adversarial examples and can attack not only classification models but also retrieval models. We conduct extensive experiments on ImageNet and CelebA datasets, validating the performance of ScoreAdv across ten target models in both black-box and white-box settings. Our results demonstrate that ScoreAdv achieves state-of-the-art attack success rates and image quality, while maintaining inference efficiency. Furthermore, the dynamic balance between denoising and adversarial perturbation enables ScoreAdv to remain robust even under defensive measures.

♻ ☆ GSE: Group-wise Sparse and Explainable Adversarial Attacks

Shpresim Sadiku, Moritz Wagner, Sebastian Pokutta

Sparse adversarial attacks fool deep neural networks (DNNs) through minimal pixel perturbations, often regularized by the $\ell_0$ norm. Recent efforts have replaced this norm with a structural sparsity regularizer, such as the nuclear group norm, to craft group-wise sparse adversarial attacks. The resulting perturbations are thus explainable and hold significant practical relevance, shedding light on an even greater vulnerability of DNNs. However, crafting such attacks poses an optimization challenge, as it involves computing norms for groups of pixels within a non-convex objective. We address this by presenting a two-phase algorithm that generates group-wise sparse attacks within semantically meaningful areas of an image. Initially, we optimize a quasinorm adversarial loss using the $1/2-$quasinorm proximal operator tailored for non-convex programming. Subsequently, the algorithm transitions to a projected Nesterov's accelerated gradient descent with $2-$norm regularization applied to perturbation magnitudes. Rigorous evaluations on CIFAR-10 and ImageNet datasets demonstrate a remarkable increase in group-wise sparsity, e.g., $50.9\%$ on CIFAR-10 and $38.4\%$ on ImageNet (average case, targeted attack). This performance improvement is accompanied by significantly faster computation times, improved explainability, and a $100\%$ attack success rate.

♻ ☆ Resource Efficient Multi-stain Kidney Glomeruli Segmentation via Self-supervision

Zeeshan Nisar, Friedrich Feuerhake, Thomas Lampert

Semantic segmentation under domain shift remains a fundamental challenge in computer vision, particularly when labelled training data is scarce. This challenge is particularly exemplified in histopathology image analysis, where the same tissue structures must be segmented across images captured under different imaging conditions (stains), each representing a distinct visual domain. Traditional deep learning methods like UNet require extensive labels, which is both costly and time-consuming, particularly when dealing with multiple domains (or stains). To mitigate this, various unsupervised domain adaptation based methods such as UDAGAN have been proposed, which reduce the need for labels by requiring only one (source) stain to be labelled. Nonetheless, obtaining source stain labels can still be challenging. This article shows that through self-supervised pre-training -- including SimCLR, BYOL, and a novel approach, HR-CS-CO -- the performance of these segmentation methods (UNet, and UDAGAN) can be retained even with 95% fewer labels. Notably, with self-supervised pre-training and using only 5% labels, the performance drops are minimal: 5.9% for UNet and 6.2% for UDAGAN, averaged over all stains, compared to their respective fully supervised counterparts (without pre-training, using 100% labels). Furthermore, these findings are shown to generalise beyond their training distribution to public benchmark datasets. Implementations and pre-trained models are publicly available \href{https://github.com/zeeshannisar/resource-effecient-multi-stain-kidney-glomeruli-segmentation.git}{online}.

comment: 39 pages, 10 figures, 4 Tables

♻ ☆ CronusVLA: Towards Efficient and Robust Manipulation via Multi-Frame Vision-Language-Action Modeling

Hao Li, Shuai Yang, Yilun Chen, Xinyi Chen, Xiaoda Yang, Yang Tian, Hanqing Wang, Tai Wang, Dahua Lin, Feng Zhao, Jiangmiao Pang

Recent vision-language-action (VLA) models built on pretrained vision-language models (VLMs) have demonstrated strong performance in robotic manipulation. However, these models remain constrained by the single-frame image paradigm and fail to fully leverage the temporal information offered by multi-frame histories, as directly feeding multiple frames into VLM backbones incurs substantial computational overhead and inference latency. We propose CronusVLA, a unified framework that extends single-frame VLA models to the multi-frame paradigm. CronusVLA follows a two-stage process: (1) Single-frame pretraining on large-scale embodied datasets with autoregressive prediction of action tokens, establishing an effective embodied vision-language foundation; (2) Multi-frame post-training, which adapts the prediction of the vision-language backbone from discrete tokens to learnable features, and aggregates historical information via feature chunking. CronusVLA effectively addresses the existing challenges of multi-frame modeling while enhancing performance and observational robustness. To evaluate the robustness under temporal and spatial disturbances, we introduce SimplerEnv-OR, a novel benchmark featuring 24 types of observational disturbances and 120 severity levels. Experiments across three embodiments in simulated and real-world environments demonstrate that CronusVLA achieves leading performance and superior robustness, with a 70.9% success rate on SimplerEnv, a 26.8% improvement over OpenVLA on LIBERO, and the highest robustness score on SimplerEnv-OR. These results highlight the potential of efficient multi-frame adaptation in VLA models for more powerful and robust real-world deployment.

comment: 39 pages, 24 figures

♻ ☆ Fit for Purpose? Deepfake Detection in the Real World

Guangyu Lin, Li Lin, Christina P. Walker, Daniel S. Schiff, Shu Hu

The rapid proliferation of AI-generated content, driven by advances in generative adversarial networks, diffusion models, and multimodal large language models, has made the creation and dissemination of synthetic media effortless, heightening the risks of misinformation, particularly political deepfakes that distort truth and undermine trust in political institutions. In turn, governments, research institutions, and industry have strongly promoted deepfake detection initiatives as solutions. Yet, most existing models are trained and validated on synthetic, laboratory-controlled datasets, limiting their generalizability to the kinds of real-world political deepfakes circulating on social platforms that affect the public. In this work, we introduce the first systematic benchmark based on the Political Deepfakes Incident Database, a curated collection of real-world political deepfakes shared on social media since 2018. Our study includes a systematic evaluation of state-of-the-art deepfake detectors across academia, government, and industry. We find that the detectors from academia and government perform relatively poorly. While paid detection tools achieve relatively higher performance than free-access models, all evaluated detectors struggle to generalize effectively to authentic political deepfakes, and are vulnerable to simple manipulations, especially in the video domain. Results urge the need for politically contextualized deepfake detection frameworks to better safeguard the public in real-world settings.

♻ ☆ DDL: A Large-Scale Datasets for Deepfake Detection and Localization in Diversified Real-World Scenarios

Changtao Miao, Yi Zhang, Weize Gao, Zhiya Tan, Weiwei Feng, Man Luo, Jianshu Li, Ajian Liu, Yunfeng Diao, Qi Chu, Tao Gong, Zhe Li, Weibin Yao, Joey Tianyi Zhou

Recent advances in AIGC have exacerbated the misuse of malicious deepfake content, making the development of reliable deepfake detection methods an essential means to address this challenge. Although existing deepfake detection models demonstrate outstanding performance in detection metrics, most methods only provide simple binary classification results, lacking interpretability. Recent studies have attempted to enhance the interpretability of classification results by providing spatial manipulation masks or temporal forgery segments. However, due to the limitations of forgery datasets, the practical effectiveness of these methods remains suboptimal. The primary reason lies in the fact that most existing deepfake datasets contain only binary labels, with limited variety in forgery scenarios, insufficient diversity in deepfake types, and relatively small data scales, making them inadequate for complex real-world scenarios.To address this predicament, we construct a novel large-scale deepfake detection and localization (\textbf{DDL}) dataset containing over $\textbf{1.4M+}$ forged samples and encompassing up to $\textbf{80}$ distinct deepfake methods. The DDL design incorporates four key innovations: (1) \textbf{Comprehensive Deepfake Methods} (covering 7 different generation architectures and a total of 80 methods), (2) \textbf{Varied Manipulation Modes} (incorporating 7 classic and 3 novel forgery modes), (3) \textbf{Diverse Forgery Scenarios and Modalities} (including 3 scenarios and 3 modalities), and (4) \textbf{Fine-grained Forgery Annotations} (providing 1.18M+ precise spatial masks and 0.23M+ precise temporal segments).Through these improvements, our DDL not only provides a more challenging benchmark for complex real-world forgeries but also offers crucial support for building next-generation deepfake detection, localization, and interpretability methods.

comment: This paper is a preliminary version, with an extended and comprehensive version currently under development

♻ ☆ HM-Talker: Hybrid Motion Modeling for High-Fidelity Talking Head Synthesis

Shiyu Liu, Kui Jiang, Xianming Liu, Hongxun Yao, Xiaocheng Feng

Audio-driven talking head video generation enhances user engagement in human-computer interaction. However, current methods frequently produce videos with motion blur and lip jitter, primarily due to their reliance on implicit modeling of audio-facial motion correlations--an approach lacking explicit articulatory priors (i.e., anatomical guidance for speech-related facial movements). To overcome this limitation, we propose HM-Talker, a novel framework for generating high-fidelity, temporally coherent talking heads. HM-Talker leverages a hybrid motion representation combining both implicit and explicit motion cues. Explicit cues use Action Units (AUs), anatomically defined facial muscle movements, alongside implicit features to minimize phoneme-viseme misalignment. Specifically, our Cross-Modal Disentanglement Module (CMDM) extracts complementary implicit/explicit motion features while predicting AUs directly from audio input aligned to visual cues. To mitigate identity-dependent biases in explicit features and enhance cross-subject generalization, we introduce the Hybrid Motion Modeling Module (HMMM). This module dynamically merges randomly paired implicit/explicit features, enforcing identity-agnostic learning. Together, these components enable robust lip synchronization across diverse identities, advancing personalized talking head synthesis. Extensive experiments demonstrate HM-Talker's superiority over state-of-the-art methods in visual quality and lip-sync accuracy.

♻ ☆ MaskCaptioner: Learning to Jointly Segment and Caption Object Trajectories in Videos

Gabriel Fiastre, Antoine Yang, Cordelia Schmid

Dense Video Object Captioning (DVOC) is the task of jointly detecting, tracking, and captioning object trajectories in a video, requiring the ability to understand spatio-temporal details and describe them in natural language. Due to the complexity of the task and the high cost associated with manual annotation, previous approaches resort to disjoint training strategies, potentially leading to suboptimal performance. To circumvent this issue, we propose to generate captions about spatio-temporally localized entities leveraging a state-of-the-art VLM. By extending the LVIS and LV-VIS datasets with our synthetic captions (LVISCap and LV-VISCap), we train MaskCaptioner, an end-to-end model capable of jointly detecting, segmenting, tracking and captioning object trajectories. Moreover, with pretraining on LVISCap and LV-VISCap, MaskCaptioner achieves state-of-the-art DVOC results on three existing benchmarks, VidSTG, VLN and BenSMOT. The datasets and code are available at https://www.gabriel.fiastre.fr/maskcaptioner/.

comment: 20 pages, 8 figures

♻ ☆ LinearSR: Unlocking Linear Attention for Stable and Efficient Image Super-Resolution

Xiaohui Li, Shaobin Zhuang, Shuo Cao, Yang Yang, Yuandong Pu, Qi Qin, Siqi Luo, Bin Fu, Yihao Liu

Generative models for Image Super-Resolution (SR) are increasingly powerful, yet their reliance on self-attention's quadratic complexity (O(N^2)) creates a major computational bottleneck. Linear Attention offers an O(N) solution, but its promise for photorealistic SR has remained largely untapped, historically hindered by a cascade of interrelated and previously unsolved challenges. This paper introduces LinearSR, a holistic framework that, for the first time, systematically overcomes these critical hurdles. Specifically, we resolve a fundamental, training instability that causes catastrophic model divergence using our novel "knee point"-based Early-Stopping Guided Fine-tuning (ESGF) strategy. Furthermore, we mitigate the classic perception-distortion trade-off with a dedicated SNR-based Mixture of Experts (MoE) architecture. Finally, we establish an effective and lightweight guidance paradigm, TAG, derived from our "precision-over-volume" principle. Our resulting LinearSR model simultaneously delivers state-of-the-art perceptual quality with exceptional efficiency. Its core diffusion forward pass (1-NFE) achieves SOTA-level speed, while its overall multi-step inference time remains highly competitive. This work provides the first robust methodology for applying Linear Attention in the photorealistic SR domain, establishing a foundational paradigm for future research in efficient generative super-resolution.

comment: 19 pages, 9 figures, 6 tables

♻ ☆ UV-Attack: Physical-World Adversarial Attacks for Person Detection via Dynamic-NeRF-based UV Mapping ICLR2025

Yanjie Li, Kaisheng Liang, Bin Xiao

In recent research, adversarial attacks on person detectors using patches or static 3D model-based texture modifications have struggled with low success rates due to the flexible nature of human movement. Modeling the 3D deformations caused by various actions has been a major challenge. Fortunately, advancements in Neural Radiance Fields (NeRF) for dynamic human modeling offer new possibilities. In this paper, we introduce UV-Attack, a groundbreaking approach that achieves high success rates even with extensive and unseen human actions. We address the challenge above by leveraging dynamic-NeRF-based UV mapping. UV-Attack can generate human images across diverse actions and viewpoints, and even create novel actions by sampling from the SMPL parameter space. While dynamic NeRF models are capable of modeling human bodies, modifying clothing textures is challenging because they are embedded in neural network parameters. To tackle this, UV-Attack generates UV maps instead of RGB images and modifies the texture stacks. This approach enables real-time texture edits and makes the attack more practical. We also propose a novel Expectation over Pose Transformation loss (EoPT) to improve the evasion success rate on unseen poses and views. Our experiments show that UV-Attack achieves a 92.7% attack success rate against the FastRCNN model across varied poses in dynamic video settings, significantly outperforming the state-of-the-art AdvCamou attack, which only had a 28.5% ASR. Moreover, we achieve 49.5% ASR on the latest YOLOv8 detector in black-box settings. This work highlights the potential of dynamic NeRF-based UV mapping for creating more effective adversarial attacks on person detectors, addressing key challenges in modeling human movement and texture modification. The code is available at https://github.com/PolyLiYJ/UV-Attack.

comment: 23 pages, 22 figures, accepted by ICLR2025

♻ ☆ A Continuous and Interpretable Morphometric for Robust Quantification of Dynamic Biological Shapes

Roua Rouatbi, Juan-Esteban Suarez Cardona, Alba Villaronga-Luque, Jesse V. Veenvliet, Ivo F. Sbalzarini

We introduce the Push-Forward Signed Distance Morphometric (PF-SDM) for shape quantification in biomedical imaging. The PF-SDM compactly encodes geometric and topological properties of closed shapes, including their skeleton and symmetries. This provides robust and interpretable features for shape comparison and machine learning. The PF-SDM is mathematically smooth, providing access to gradients and differential-geometric quantities. It also extends to temporal dynamics and allows fusing spatial intensity distributions, such as genetic markers, with shape dynamics. We present the PF-SDM theory, benchmark it on synthetic data, and apply it to predicting body-axis formation in mouse gastruloids, outperforming a CNN baseline in both accuracy and speed.

♻ ☆ VerifIoU -- Robustness of Object Detection to Perturbations SC

Noémie Cohen, Mélanie Ducoffe, Ryma Boumazouza, Christophe Gabreau, Claire Pagetti, Xavier Pucel, Audrey Galametz

We introduce a novel Interval Bound Propagation (IBP) approach for the formal verification of object detection models, specifically targeting the Intersection over Union (IoU) metric. The approach has been implemented in an open source code, named IBP IoU, compatible with popular abstract interpretation based verification tools. The resulting verifier is evaluated on landing approach runway detection and handwritten digit recognition case studies. Comparisons against a baseline (Vanilla IBP IoU) highlight the superior performance of IBP IoU in ensuring accuracy and stability, contributing to more secure and robust machine learning applications.

comment: 44th Digital Avionics Systems Conference (DASC), Sep 2025, Montreal, Canada

♻ ☆ ReCon-GS: Continuum-Preserved Gaussian Streaming for Fast and Compact Reconstruction of Dynamic Scenes NeurIPS 2025

Jiaye Fu, Qiankun Gao, Chengxiang Wen, Yanmin Wu, Siwei Ma, Jiaqi Zhang, Jian Zhang

Online free-viewpoint video (FVV) reconstruction is challenged by slow per-frame optimization, inconsistent motion estimation, and unsustainable storage demands. To address these challenges, we propose the Reconfigurable Continuum Gaussian Stream, dubbed ReCon-GS, a novel storage-aware framework that enables high fidelity online dynamic scene reconstruction and real-time rendering. Specifically, we dynamically allocate multi-level Anchor Gaussians in a density-adaptive fashion to capture inter-frame geometric deformations, thereby decomposing scene motion into compact coarse-to-fine representations. Then, we design a dynamic hierarchy reconfiguration strategy that preserves localized motion expressiveness through on-demand anchor re-hierarchization, while ensuring temporal consistency through intra-hierarchical deformation inheritance that confines transformation priors to their respective hierarchy levels. Furthermore, we introduce a storage-aware optimization mechanism that flexibly adjusts the density of Anchor Gaussians at different hierarchy levels, enabling a controllable trade-off between reconstruction fidelity and memory usage. Extensive experiments on three widely used datasets demonstrate that, compared to state-of-the-art methods, ReCon-GS improves training efficiency by approximately 15% and achieves superior FVV synthesis quality with enhanced robustness and stability. Moreover, at equivalent rendering quality, ReCon-GS slashes memory requirements by over 50% compared to leading state-of-the-art methods.

comment: Published in NeurIPS 2025

♻ ☆ StyleGuard: Preventing Text-to-Image-Model-based Style Mimicry Attacks by Style Perturbations NIPS2025

Yanjie Li, Wenxuan Zhang, Xinqi Lyu, Yihao Liu, Bin Xiao

Recently, text-to-image diffusion models have been widely used for style mimicry and personalized customization through methods such as DreamBooth and Textual Inversion. This has raised concerns about intellectual property protection and the generation of deceptive content. Recent studies, such as Glaze and Anti-DreamBooth, have proposed using adversarial noise to protect images from these attacks. However, recent purification-based methods, such as DiffPure and Noise Upscaling, have successfully attacked these latest defenses, showing the vulnerabilities of these methods. Moreover, present methods show limited transferability across models, making them less effective against unknown text-to-image models. To address these issues, we propose a novel anti-mimicry method, StyleGuard. We propose a novel style loss that optimizes the style-related features in the latent space to make it deviate from the original image, which improves model-agnostic transferability. Additionally, to enhance the perturbation's ability to bypass diffusion-based purification, we designed a novel upscale loss that involves ensemble purifiers and upscalers during training. Extensive experiments on the WikiArt and CelebA datasets demonstrate that StyleGuard outperforms existing methods in robustness against various transformations and purifications, effectively countering style mimicry in various models. Moreover, StyleGuard is effective on different style mimicry methods, including DreamBooth and Textual Inversion. The code is available at https://github.com/PolyLiYJ/StyleGuard.

comment: Accepted by NIPS2025

♻ ☆ LATex: Leveraging Attribute-based Text Knowledge for Aerial-Ground Person Re-Identification

Pingping Zhang, Xiang Hu, Yuhao Wang, Huchuan Lu

As an important task in intelligent transportation systems, Aerial-Ground person Re-IDentification (AG-ReID) aims to retrieve specific persons across heterogeneous cameras in different viewpoints. Previous methods typically adopt deep learning-based models, focusing on extracting view-invariant features. However, they usually overlook the semantic information in person attributes. In addition, existing training strategies often rely on full fine-tuning large-scale models, which significantly increases training costs. To address these issues, we propose a novel framework named LATex for AG-ReID, which adopts prompt-tuning strategies to leverage attribute-based text knowledge. Specifically, with the Contrastive Language-Image Pre-training (CLIP) model, we first propose an Attribute-aware Image Encoder (AIE) to extract both global semantic features and attribute-aware features from input images. Then, with these features, we propose a Prompted Attribute Classifier Group (PACG) to predict person attributes and obtain attribute representations. Finally, we design a Coupled Prompt Template (CPT) to transform attribute representations and view information into structured sentences. These sentences are processed by the text encoder of CLIP to generate more discriminative features. As a result, our framework can fully leverage attribute-based text knowledge to improve AG-ReID performance. Extensive experiments on three AG-ReID benchmarks demonstrate the effectiveness of our proposed methods. The source code is available at https://github.com/kevinhu314/LATex.

comment: More modifications may be performed

♻ ☆ EmoAttack: Emotion-to-Image Diffusion Models for Emotional Backdoor Generation

Tianyu Wei, Shanmin Pang, Qi Guo, Yizhuo Ma, Xiaofeng Cao, Qing Guo

Text-to-image diffusion models can generate realistic images based on textual inputs, enabling users to convey their opinions visually through language. Meanwhile, within language, emotion plays a crucial role in expressing personal opinions in our daily lives and the inclusion of maliciously negative content can lead users astray, exacerbating negative emotions. Recognizing the success of diffusion models and the significance of emotion, we investigate a previously overlooked risk associated with text-to-image diffusion models, that is, utilizing emotion in the input texts to introduce negative content and provoke unfavorable emotions in users. Specifically, we identify a new backdoor attack, i.e., emotion-aware backdoor attack (EmoAttack), which introduces malicious negative content triggered by emotional texts during image generation. We formulate such an attack as a diffusion personalization problem to avoid extensive model retraining and propose the EmoBooth. Unlike existing personalization methods, our approach fine-tunes a pre-trained diffusion model by establishing a mapping between a cluster of emotional words and a given reference image containing malicious negative content. To validate the effectiveness of our method, we built a dataset and conducted extensive analysis and discussion about its effectiveness. Given consumers' widespread use of diffusion models, uncovering this threat is critical for society.

♻ ☆ Two Heads are Better than One: Robust Learning Meets Multi-branch Models

Zongyuan Zhang, Qingwen Bu, Tianyang Duan, Zheng Lin, Yuhao Qing, Zihan Fang, Heming Cui, Dong Huang

Deep neural networks (DNNs) are vulnerable to adversarial examples, in which DNNs are misled to false outputs due to inputs containing imperceptible perturbations. Adversarial training, a reliable and effective method of defense, may significantly reduce the vulnerability of neural networks and becomes the de facto standard for robust learning. While many recent works practice the data-centric philosophy, such as how to generate better adversarial examples or use generative models to produce additional training data, we look back to the models themselves and revisit the adversarial robustness from the perspective of deep feature distribution as an insightful complementarity. In this paper, we propose \textit{Branch Orthogonality adveRsarial Training} (BORT) to obtain state-of-the-art performance with solely the original dataset for adversarial training. To practice our design idea of integrating multiple orthogonal solution spaces, we leverage a simple and straightforward multi-branch neural network that eclipses adversarial attacks with no increase in inference time. We heuristically propose a corresponding loss function, branch-orthogonal loss, to make each solution space of the multi-branch model orthogonal. We evaluate our approach on CIFAR-10, CIFAR-100 and SVHN against $\ell_{\infty}$ norm-bounded perturbations of size $\epsilon = 8/255$, respectively. Exhaustive experiments are conducted to show that our method goes beyond all state-of-the-art methods without any tricks. Compared to all methods that do not use additional data for training, our models achieve 67.3\% and 41.5\% robust accuracy on CIFAR-10 and CIFAR-100 (improving upon the state-of-the-art by +7.23\% and +9.07\%). We also outperform methods using a training set with a far larger scale than ours.

comment: Camera-ready version for ICPADS 2025

♻ ☆ Static for Dynamic: Towards a Deeper Understanding of Dynamic Facial Expressions Using Static Expression Data

Yin Chen, Jia Li, Yu Zhang, Zhenzhen Hu, Shiguang Shan, Meng Wang, Richang Hong

Dynamic facial expression recognition (DFER) infers emotions from the temporal evolution of expressions, unlike static facial expression recognition (SFER), which relies solely on a single snapshot. This temporal analysis provides richer information and promises greater recognition capability. However, current DFER methods often exhibit unsatisfied performance largely due to fewer training samples compared to SFER. Given the inherent correlation between static and dynamic expressions, we hypothesize that leveraging the abundant SFER data can enhance DFER. To this end, we propose Static-for-Dynamic (S4D), a unified dual-modal learning framework that integrates SFER data as a complementary resource for DFER. Specifically, S4D employs dual-modal self-supervised pre-training on facial images and videos using a shared Vision Transformer (ViT) encoder-decoder architecture, yielding improved spatiotemporal representations. The pre-trained encoder is then fine-tuned on static and dynamic expression datasets in a multi-task learning setup to facilitate emotional information interaction. Unfortunately, vanilla multi-task learning in our study results in negative transfer. To address this, we propose an innovative Mixture of Adapter Experts (MoAE) module that facilitates task-specific knowledge acquisition while effectively extracting shared knowledge from both static and dynamic expression data. Extensive experiments demonstrate that S4D achieves a deeper understanding of DFER, setting new state-of-the-art performance on FERV39K, MAFW, and DFEW benchmarks, with weighted average recall (WAR) of 53.65\%, 58.44\%, and 76.68\%, respectively. Additionally, a systematic correlation analysis between SFER and DFER tasks is presented, which further elucidates the potential benefits of leveraging SFER.

comment: The code and model are publicly available here https://github.com/MSA-LMC/S4D

♻ ☆ Tunable-Generalization Diffusion Powered by Self-Supervised Contextual Sub-Data for Low-Dose CT Reconstruction

Guoquan Wei, Liu Shi, Zekun Zhou, Wenzhe Shan, Qiegen Liu

Current models based on deep learning for low-dose CT denoising rely heavily on paired data and generalize poorly. Even the more concerned diffusion models need to learn the distribution of clean data for reconstruction, which is difficult to satisfy in medical clinical applications. At the same time, self-supervised-based methods face the challenge of significant degradation of generalizability of models pre-trained for the current dose to expand to other doses. To address these issues, this work proposes a novel method of TUnable-geneRalizatioN Diffusion (TurnDiff) powered by self-supervised contextual sub-data for low-dose CT reconstruction. Firstly, a contextual subdata self-enhancing similarity strategy is designed for denoising centered on the LDCT projection domain, which provides an initial prior for the subsequent progress. Subsequently, the initial prior is used to combine knowledge distillation with a deep combination of latent diffusion models for optimizing image details. The pre-trained model is used for inference reconstruction, and the pixel-level self-correcting fusion technique is proposed for fine-grained reconstruction of the image domain to enhance the image fidelity, using the initial prior and the LDCT image as a guide. In addition, the technique is flexibly applied to the generalization of upper and lower doses or even unseen doses. Dual-domain strategy cascade for self-supervised LDCT denoising, TurnDiff requires only LDCT projection domain data for training and testing. Comprehensive evaluation on both benchmark datasets and real-world data demonstrates that TurnDiff consistently outperforms state-of-the-art methods in both reconstruction and generalization.

♻ ☆ Seeing Structural Failure Before it Happens: An Image-Based Physics-Informed Neural Network (PINN) for Spaghetti Bridge Load Prediction

Omer Jauhar Khan, Sudais Khan, Hafeez Anwar, Shahzeb Khan, Shams Ul Arifeen

Physics Informed Neural Networks (PINNs) are gaining attention for their ability to embed physical laws into deep learning models, which is particularly useful in structural engineering tasks with limited data. This paper aims to explore the use of PINNs to predict the weight of small scale spaghetti bridges, a task relevant to understanding load limits and potential failure modes in simplified structural models. Our proposed framework incorporates physics-based constraints to the prediction model for improved performance. In addition to standard PINNs, we introduce a novel architecture named Physics Informed Kolmogorov Arnold Network (PIKAN), which blends universal function approximation theory with physical insights. The structural parameters provided as input to the model are collected either manually or through computer vision methods. Our dataset includes 15 real bridges, augmented to 100 samples, and our best model achieves an $R^2$ score of 0.9603 and a mean absolute error (MAE) of 10.50 units. From applied perspective, we also provide a web based interface for parameter entry and prediction. These results show that PINNs can offer reliable estimates of structural weight, even with limited data, and may help inform early stage failure analysis in lightweight bridge designs. The complete data and code are available at https://github.com/OmerJauhar/PINNS-For-Spaghetti-Bridges.

comment: 12 pages, 17 figures. Preprint

♻ ☆ SD-ReID: View-aware Stable Diffusion for Aerial-Ground Person Re-Identification

Yuhao Wang, Xiang Hu, Lixin Wang, Pingping Zhang, Huchuan Lu

Aerial-Ground Person Re-IDentification (AG-ReID) aims to retrieve specific persons across cameras with different viewpoints. Previous works focus on designing discriminative models to maintain the identity consistency despite drastic changes in camera viewpoints. The core idea behind these methods is quite natural, but designing a view-robust model is a very challenging task. Moreover, they overlook the contribution of view-specific features in enhancing the model's ability to represent persons. To address these issues, we propose a novel generative framework named SD-ReID for AG-ReID, which leverages generative models to mimic the feature distribution of different views while extracting robust identity representations. More specifically, we first train a ViT-based model to extract person representations along with controllable conditions, including identity and view conditions. We then fine-tune the Stable Diffusion (SD) model to enhance person representations guided by these controllable conditions. Furthermore, we introduce the View-Refined Decoder (VRD) to bridge the gap between instance-level and global-level features. Finally, both person representations and all-view features are employed to retrieve target persons. Extensive experiments on five AG-ReID benchmarks (i.e., CARGO, AG-ReIDv1, AG-ReIDv2, LAGPeR and G2APS-ReID) demonstrate the effectiveness of our proposed method. The source code will be available.

comment: More modifications may performed

♻ ☆ TRUST-VL: An Explainable News Assistant for General Multimodal Misinformation Detection EMNLP 2025

Zehong Yan, Peng Qi, Wynne Hsu, Mong Li Lee

Multimodal misinformation, encompassing textual, visual, and cross-modal distortions, poses an increasing societal threat that is amplified by generative AI. Existing methods typically focus on a single type of distortion and struggle to generalize to unseen scenarios. In this work, we observe that different distortion types share common reasoning capabilities while also requiring task-specific skills. We hypothesize that joint training across distortion types facilitates knowledge sharing and enhances the model's ability to generalize. To this end, we introduce TRUST-VL, a unified and explainable vision-language model for general multimodal misinformation detection. TRUST-VL incorporates a novel Question-Aware Visual Amplifier module, designed to extract task-specific visual features. To support training, we also construct TRUST-Instruct, a large-scale instruction dataset containing 198K samples featuring structured reasoning chains aligned with human fact-checking workflows. Extensive experiments on both in-domain and zero-shot benchmarks demonstrate that TRUST-VL achieves state-of-the-art performance, while also offering strong generalization and interpretability.

comment: EMNLP 2025 Oral; Project Homepage: https://yanzehong.github.io/trust-vl/

♻ ☆ Paper2Poster: Towards Multimodal Poster Automation from Scientific Papers

Wei Pang, Kevin Qinghong Lin, Xiangru Jian, Xi He, Philip Torr

Academic poster generation is a crucial yet challenging task in scientific communication, requiring the compression of long-context interleaved documents into a single, visually coherent page. To address this challenge, we introduce the first benchmark and metric suite for poster generation, which pairs recent conference papers with author-designed posters and evaluates outputs on (i)Visual Quality-semantic alignment with human posters, (ii)Textual Coherence-language fluency, (iii)Holistic Assessment-six fine-grained aesthetic and informational criteria scored by a VLM-as-judge, and notably (iv)PaperQuiz-the poster's ability to convey core paper content as measured by VLMs answering generated quizzes. Building on this benchmark, we propose PosterAgent, a top-down, visual-in-the-loop multi-agent pipeline: the (a)Parser distills the paper into a structured asset library; the (b)Planner aligns text-visual pairs into a binary-tree layout that preserves reading order and spatial balance; and the (c)Painter-Commenter loop refines each panel by executing rendering code and using VLM feedback to eliminate overflow and ensure alignment. In our comprehensive evaluation, we find that GPT-4o outputs-though visually appealing at first glance-often exhibit noisy text and poor PaperQuiz scores, and we find that reader engagement is the primary aesthetic bottleneck, as human-designed posters rely largely on visual semantics to convey meaning. Our fully open-source variants (e.g. based on the Qwen-2.5 series) outperform existing 4o-driven multi-agent systems across nearly all metrics, while using 87% fewer tokens. It transforms a 22-page paper into a finalized yet editable .pptx poster - all for just $0.005. These findings chart clear directions for the next generation of fully automated poster-generation models. The code and datasets are available at https://github.com/Paper2Poster/Paper2Poster.

comment: Project Page: https://github.com/Paper2Poster/Paper2Poster

♻ ☆ MindGYM: What Matters in Question Synthesis for Thinking-Centric Fine-Tuning? NeurIPS'25

Zhe Xu, Daoyuan Chen, Zhenqing Ling, Yaliang Li, Ying Shen

Large foundation models face challenges in acquiring transferable, structured thinking abilities, especially when supervised with rigid templates or crowd-annotated instruction datasets. Unlike prior approaches, we focus on a thinking-centric data synthesis paradigm that enables models to evolve through self-generated, cognitively guided data. We propose MindGYM, a structured and scalable framework for question synthesis, composed of: (1) Cognitive Thinking Process Injection, which infuses high-level reasoning objectives to shape the model's synthesis behavior; (2) Seed Single-Hop Question Synthesis, generating atomic questions from diverse semantic types to encourage broader thinking; and (3) Challenging Multi-Hop QA Synthesis, composing more complex multi-hop questions based on QA seeds for deeper reasoning. Detailed analysis shows that synthetic data generated by our method achieves 16.7% higher average quality and 67.91% lower quality variance compared to baseline sources, highlighting that both high-quality and self-contained data are essential for effective, thinking-oriented fine-tuning. MindGYM improves performance on six reasoning benchmarks, achieving gains of up to 16% on MathVision using only 400 data samples, and generalizable improvements across different model sizes and architectures. MindGYM underscores the viability of self-challenging mechanisms in refining large model capabilities while minimizing human intervention and resource demands. Code and data are released to promote data-centric research into self-evolving foundation models driven by their internal reasoning capabilities.

comment: Accepted by NeurIPS'25. 30 pages, 2 figures, 13 tables

♻ ☆ D-HUMOR: Dark Humor Understanding via Multimodal Open-ended Reasoning -- A Benchmark Dataset and Method ICDM

Sai Kartheek Reddy Kasu, Mohammad Zia Ur Rehman, Shahid Shafi Dar, Rishi Bharat Junghare, Dhanvin Sanjay Namboodiri, Nagendra Kumar

Dark humor in online memes poses unique challenges due to its reliance on implicit, sensitive, and culturally contextual cues. To address the lack of resources and methods for detecting dark humor in multimodal content, we introduce a novel dataset of 4,379 Reddit memes annotated for dark humor, target category (gender, mental health, violence, race, disability, and other), and a three-level intensity rating (mild, moderate, severe). Building on this resource, we propose a reasoning-augmented framework that first generates structured explanations for each meme using a Large Vision-Language Model (VLM). Through a Role-Reversal Self-Loop, VLM adopts the author's perspective to iteratively refine its explanations, ensuring completeness and alignment. We then extract textual features from both the OCR transcript and the self-refined reasoning via a text encoder, while visual features are obtained using a vision transformer. A Tri-stream Cross-Reasoning Network (TCRNet) fuses these three streams, text, image, and reasoning, via pairwise attention mechanisms, producing a unified representation for classification. Experimental results demonstrate that our approach outperforms strong baselines across three tasks: dark humor detection, target identification, and intensity prediction. The dataset, annotations, and code are released to facilitate further research in multimodal humor understanding and content moderation. Code and Dataset are available at: https://github.com/Sai-Kartheek-Reddy/D-Humor-Dark-Humor-Understanding-via-Multimodal-Open-ended-Reasoning

comment: Accepted at IEEE International Conference on Data Mining (ICDM) 2025

♻ ☆ Disentangled 4D Gaussian Splatting: Rendering High-Resolution Dynamic World at 343 FPS

Hao Feng, Hao Sun, Wei Xie, Zhi Zuo, Zhengzhe Liu

While dynamic novel view synthesis from 2D videos has seen progress, achieving efficient reconstruction and rendering of dynamic scenes remains a challenging task. In this paper, we introduce Disentangled 4D Gaussian Splatting (Disentangled4DGS), a novel representation and rendering pipeline that achieves real-time performance without compromising visual fidelity. Disentangled4DGS decouples the temporal and spatial components of 4D Gaussians, avoiding the need for slicing first and four-dimensional matrix calculations in prior methods. By projecting temporal and spatial deformations into dynamic 2D Gaussians and deferring temporal processing, we minimize redundant computations of 4DGS. Our approach also features a gradient-guided flow loss and temporal splitting strategy to reduce artifacts. Experiments demonstrate a significant improvement in rendering speed and quality, achieving 343 FPS when render 1352*1014 resolution images on a single RTX3090 while reducing storage requirements by at least 4.5%. Our approach sets a new benchmark for dynamic novel view synthesis, outperforming existing methods on both multi-view and monocular dynamic scene datasets.

♻ ☆ GRPO-Guard: Mitigating Implicit Over-Optimization in Flow Matching via Regulated Clipping

Jing Wang, Jiajun Liang, Jie Liu, Henglin Liu, Gongye Liu, Jun Zheng, Wanyuan Pang, Ao Ma, Zhenyu Xie, Xintao Wang, Meng Wang, Pengfei Wan, Xiaodan Liang

Recently, GRPO-based reinforcement learning has shown remarkable progress in optimizing flow-matching models, effectively improving their alignment with task-specific rewards. Within these frameworks, the policy update relies on importance-ratio clipping to constrain overconfident positive and negative gradients. However, in practice, we observe a systematic shift in the importance-ratio distribution-its mean falls below 1 and its variance differs substantially across timesteps. This left-shifted and inconsistent distribution prevents positive-advantage samples from entering the clipped region, causing the mechanism to fail in constraining overconfident positive updates. As a result, the policy model inevitably enters an implicit over-optimization stage-while the proxy reward continues to increase, essential metrics such as image quality and text-prompt alignment deteriorate sharply, ultimately making the learned policy impractical for real-world use. To address this issue, we introduce GRPO-Guard, a simple yet effective enhancement to existing GRPO frameworks. Our method incorporates ratio normalization, which restores a balanced and step-consistent importance ratio, ensuring that PPO clipping properly constrains harmful updates across denoising timesteps. In addition, a gradient reweighting strategy equalizes policy gradients over noise conditions, preventing excessive updates from particular timestep regions. Together, these designs act as a regulated clipping mechanism, stabilizing optimization and substantially mitigating implicit over-optimization without relying on heavy KL regularization. Extensive experiments on multiple diffusion backbones (e.g., SD3.5M, Flux.1-dev) and diverse proxy tasks demonstrate that GRPO-Guard significantly reduces over-optimization while maintaining or even improving generation quality.

comment: Project Page: https://jingw193.github.io/GRPO-Guard/

♻ ☆ RRCANet: Recurrent Reusable-Convolution Attention Network for Infrared Small Target Detection

Yongxian Liu, Boyang Li, Ting Liu, Zaiping Lin, Wei An

Infrared small target detection is a challenging task due to its unique characteristics (e.g., small, dim, shapeless and changeable). Recently published CNN-based methods have achieved promising performance with heavy feature extraction and fusion modules. To achieve efficient and effective detection, we propose a recurrent reusable-convolution attention network (RRCA-Net) for infrared small target detection. Specifically, RRCA-Net incorporates reusable-convolution block (RuCB) in a recurrent manner without introducing extra parameters. With the help of the repetitive iteration in RuCB, the high-level information of small targets in the deep layers can be well maintained and further refined. Then, a dual interactive attention aggregation module (DIAAM) is proposed to promote the mutual enhancement and fusion of refined information. In this way, RRCA-Net can both achieve high-level feature refinement and enhance the correlation of contextual information between adjacent layers. Moreover, to achieve steady convergence, we design a target characteristic inspired loss function (DpT-k loss) by integrating physical and mathematical constraints. Experimental results on three benchmark datasets (e.g. NUAA-SIRST, IRSTD-1k, DenseSIRST) demonstrate that our RRCA-Net can achieve comparable performance to the state-of-the-art methods while maintaining a small number of parameters, and act as a plug and play module to introduce consistent performance improvement for several popular IRSTD methods.

comment: We have updated the journal reference and DOI

♻ ☆ Open3D-VQA: A Benchmark for Comprehensive Spatial Reasoning with Multimodal Large Language Model in Open Space

Weichen Zhang, Zile Zhou, Xin Zeng, Xuchen Liu, Jianjie Fang, Chen Gao, Yong Li, Jinqiang Cui, Xinlei Chen, Xiao-Ping Zhang

Spatial reasoning is a fundamental capability of multimodal large language models (MLLMs), yet their performance in open aerial environments remains underexplored. In this work, we present Open3D-VQA, a novel benchmark for evaluating MLLMs' ability to reason about complex spatial relationships from an aerial perspective. The benchmark comprises 73k QA pairs spanning 7 general spatial reasoning tasks, including multiple-choice, true/false, and short-answer formats, and supports both visual and point cloud modalities. The questions are automatically generated from spatial relations extracted from both real-world and simulated aerial scenes. Evaluation on 13 popular MLLMs reveals that: 1) Models are generally better at answering questions about relative spatial relations than absolute distances, 2) 3D LLMs fail to demonstrate significant advantages over 2D LLMs, and 3) Fine-tuning solely on the simulated dataset can significantly improve the model's spatial reasoning performance in real-world scenarios. We release our benchmark, data generation pipeline, and evaluation toolkit to support further research: https://github.com/EmbodiedCity/Open3D-VQA.code.

♻ ☆ FASL-Seg: Anatomy and Tool Segmentation of Surgical Scenes ECAI

Muraam Abdel-Ghani, Mahmoud Ali, Mohamed Ali, Fatmaelzahraa Ahmed, Muhammad Arsalan, Abdulaziz Al-Ali, Shidin Balakrishnan

The growing popularity of robotic minimally invasive surgeries has made deep learning-based surgical training a key area of research. A thorough understanding of the surgical scene components is crucial, which semantic segmentation models can help achieve. However, most existing work focuses on surgical tools and overlooks anatomical objects. Additionally, current state-of-the-art (SOTA) models struggle to balance capturing high-level contextual features and low-level edge features. We propose a Feature-Adaptive Spatial Localization model (FASL-Seg), designed to capture features at multiple levels of detail through two distinct processing streams, namely a Low-Level Feature Projection (LLFP) and a High-Level Feature Projection (HLFP) stream, for varying feature resolutions - enabling precise segmentation of anatomy and surgical instruments. We evaluated FASL-Seg on surgical segmentation benchmark datasets EndoVis18 and EndoVis17 on three use cases. The FASL-Seg model achieves a mean Intersection over Union (mIoU) of 72.71% on parts and anatomy segmentation in EndoVis18, improving on SOTA by 5%. It further achieves a mIoU of 85.61% and 72.78% in EndoVis18 and EndoVis17 tool type segmentation, respectively, outperforming SOTA overall performance, with comparable per-class SOTA results in both datasets and consistent performance in various classes for anatomy and instruments, demonstrating the effectiveness of distinct processing streams for varying feature resolutions.

comment: 8 pages, 6 figures, In Proceedings of European Conference on Artificial Intelligence (ECAI) 2025

♻ ☆ Omni-Effects: Unified and Spatially-Controllable Visual Effects Generation

Fangyuan Mao, Aiming Hao, Jintao Chen, Dongxia Liu, Xiaokun Feng, Jiashu Zhu, Meiqi Wu, Chubin Chen, Jiahong Wu, Xiangxiang Chu

Visual effects (VFX) are essential visual enhancements fundamental to modern cinematic production. Although video generation models offer cost-efficient solutions for VFX production, current methods are constrained by per-effect LoRA training, which limits generation to single effects. This fundamental limitation impedes applications that require spatially controllable composite effects, i.e., the concurrent generation of multiple effects at designated locations. However, integrating diverse effects into a unified framework faces major challenges: interference from effect variations and spatial uncontrollability during multi-VFX joint training. To tackle these challenges, we propose Omni-Effects, a first unified framework capable of generating prompt-guided effects and spatially controllable composite effects. The core of our framework comprises two key innovations: (1) LoRA-based Mixture of Experts (LoRA-MoE), which employs a group of expert LoRAs, integrating diverse effects within a unified model while effectively mitigating cross-task interference. (2) Spatial-Aware Prompt (SAP) incorporates spatial mask information into the text token, enabling precise spatial control. Furthermore, we introduce an Independent-Information Flow (IIF) module integrated within the SAP, isolating the control signals corresponding to individual effects to prevent any unwanted blending. To facilitate this research, we construct a comprehensive VFX dataset Omni-VFX via a novel data collection pipeline combining image editing and First-Last Frame-to-Video (FLF2V) synthesis, and introduce a dedicated VFX evaluation framework for validating model performance. Extensive experiments demonstrate that Omni-Effects achieves precise spatial control and diverse effect generation, enabling users to specify both the category and location of desired effects.

♻ ☆ Defending Multimodal Backdoored Models by Repulsive Visual Prompt Tuning

Zhifang Zhang, Shuo He, Haobo Wang, Bingquan Shen, Lei Feng

Multimodal contrastive learning models (e.g., CLIP) can learn high-quality representations from large-scale image-text datasets, while they exhibit significant vulnerabilities to backdoor attacks, raising serious safety concerns. In this paper, we reveal that CLIP's vulnerabilities primarily stem from its tendency to encode features beyond in-dataset predictive patterns, compromising its visual feature resistivity to input perturbations. This makes its encoded features highly susceptible to being reshaped by backdoor triggers. To address this challenge, we propose Repulsive Visual Prompt Tuning (RVPT), a novel defense approach that employs deep visual prompt tuning with a specially designed feature-repelling loss. Specifically, RVPT adversarially repels the encoded features from deeper layers while optimizing the standard cross-entropy loss, ensuring that only predictive features in downstream tasks are encoded, thereby enhancing CLIP's visual feature resistivity against input perturbations and mitigating its susceptibility to backdoor attacks. Unlike existing multimodal backdoor defense methods that typically require the availability of poisoned data or involve fine-tuning the entire model, RVPT leverages few-shot downstream clean samples and only tunes a small number of parameters. Empirical results demonstrate that RVPT tunes only 0.27\% of the parameters in CLIP, yet it significantly outperforms state-of-the-art defense methods, reducing the attack success rate from 89.70\% to 2.76\% against the most advanced multimodal attacks on ImageNet and effectively generalizes its defensive capabilities across multiple datasets.

♻ ☆ FARMER: Flow AutoRegressive Transformer over Pixels

Guangting Zheng, Qinyu Zhao, Tao Yang, Fei Xiao, Zhijie Lin, Jie Wu, Jiajun Deng, Yanyong Zhang, Rui Zhu

Directly modeling the explicit likelihood of the raw data distribution is key topic in the machine learning area, which achieves the scaling successes in Large Language Models by autoregressive modeling. However, continuous AR modeling over visual pixel data suffer from extremely long sequences and high-dimensional spaces. In this paper, we present FARMER, a novel end-to-end generative framework that unifies Normalizing Flows (NF) and Autoregressive (AR) models for tractable likelihood estimation and high-quality image synthesis directly from raw pixels. FARMER employs an invertible autoregressive flow to transform images into latent sequences, whose distribution is modeled implicitly by an autoregressive model. To address the redundancy and complexity in pixel-level modeling, we propose a self-supervised dimension reduction scheme that partitions NF latent channels into informative and redundant groups, enabling more effective and efficient AR modeling. Furthermore, we design a one-step distillation scheme to significantly accelerate inference speed and introduce a resampling-based classifier-free guidance algorithm to boost image generation quality. Extensive experiments demonstrate that FARMER achieves competitive performance compared to existing pixel-based generative models while providing exact likelihoods and scalable training.

comment: Bytedance Seed Technical Report

♻ ☆ SPARKE: Scalable Prompt-Aware Diversity and Novelty Guidance in Diffusion Models via RKE Score

Mohammad Jalali, Haoyu Lei, Amin Gohari, Farzan Farnia

Diffusion models have demonstrated remarkable success in high-fidelity image synthesis and prompt-guided generative modeling. However, ensuring adequate diversity in generated samples of prompt-guided diffusion models remains a challenge, particularly when the prompts span a broad semantic spectrum and the diversity of generated data needs to be evaluated in a prompt-aware fashion across semantically similar prompts. Recent methods have introduced guidance via diversity measures to encourage more varied generations. In this work, we extend the diversity measure-based approaches by proposing the Scalable Prompt-Aware R\'eny Kernel Entropy Diversity Guidance (SPARKE) method for prompt-aware diversity guidance. SPARKE utilizes conditional entropy for diversity guidance, which dynamically conditions diversity measurement on similar prompts and enables prompt-aware diversity control. While the entropy-based guidance approach enhances prompt-aware diversity, its reliance on the matrix-based entropy scores poses computational challenges in large-scale generation settings. To address this, we focus on the special case of Conditional latent RKE Score Guidance, reducing entropy computation and gradient-based optimization complexity from the $O(n^3)$ of general entropy measures to $O(n)$. The reduced computational complexity allows for diversity-guided sampling over potentially thousands of generation rounds on different prompts. We numerically test the SPARKE method on several text-to-image diffusion models, demonstrating that the proposed method improves the prompt-aware diversity of the generated data without incurring significant computational costs. We release our code on the project page: https://mjalali.github.io/SPARKE

♻ ☆ Real-Time Neural Video Compression with Unified Intra and Inter Coding

Hui Xiang, Yifan Bian, Li Li, Jingran Wu, Xianguo Zhang, Dong Liu

Neural video compression (NVC) technologies have advanced rapidly in recent years, yielding state-of-the-art schemes such as DCVC-RT that offer superior compression efficiency to H.266/VVC and real-time encoding/decoding capabilities. Nonetheless, existing NVC schemes have several limitations, including inefficiency in dealing with disocclusion and new content, interframe error propagation and accumulation, among others. To eliminate these limitations, we borrow the idea from classic video coding schemes, which allow intra coding within inter-coded frames. With the intra coding tool enabled, disocclusion and new content are properly handled, and interframe error propagation is naturally intercepted without the need for manual refresh mechanisms. We present an NVC framework with unified intra and inter coding, where every frame is processed by a single model that is trained to perform intra/inter coding adaptively. Moreover, we propose a simultaneous two-frame compression design to exploit interframe redundancy not only forwardly but also backwardly. Experimental results show that our scheme outperforms DCVC-RT by an average of 12.1% BD-rate reduction, delivers more stable bitrate and quality per frame, and retains real-time encoding/decoding performances. Code and models will be released.

comment: 10 pages

♻ ☆ TeleEgo: Benchmarking Egocentric AI Assistants in the Wild

Jiaqi Yan, Ruilong Ren, Jingren Liu, Shuning Xu, Ling Wang, Yiheng Wang, Yun Wang, Long Zhang, Xiangyu Chen, Changzhi Sun, Jixiang Luo, Dell Zhang, Hao Sun, Chi Zhang, Xuelong Li

Egocentric AI assistants in real-world settings must process multi-modal inputs (video, audio, text), respond in real time, and retain evolving long-term memory. However, existing benchmarks typically evaluate these abilities in isolation, lack realistic streaming scenarios, or support only short-term tasks. We introduce \textbf{TeleEgo}, a long-duration, streaming, omni-modal benchmark for evaluating egocentric AI assistants in realistic daily contexts. The dataset features over 14 hours per participant of synchronized egocentric video, audio, and text across four domains: work \& study, lifestyle \& routines, social activities, and outings \& culture. All data is aligned on a unified global timeline and includes high-quality visual narrations and speech transcripts, curated through human refinement.TeleEgo defines 12 diagnostic subtasks across three core capabilities: Memory (recalling past events), Understanding (interpreting the current moment), and Cross-Memory Reasoning (linking distant events). It contains 3,291 human-verified QA items spanning multiple question formats (single-choice, binary, multi-choice, and open-ended), evaluated strictly in a streaming setting. We propose two key metrics -- Real-Time Accuracy and Memory Persistence Time -- to jointly assess correctness, temporal responsiveness, and long-term retention. TeleEgo provides a realistic and comprehensive evaluation to advance the development of practical AI assistants.

♻ ☆ DOVE: Efficient One-Step Diffusion Model for Real-World Video Super-Resolution NeurIPS 2025

Zheng Chen, Zichen Zou, Kewei Zhang, Xiongfei Su, Xin Yuan, Yong Guo, Yulun Zhang

Diffusion models have demonstrated promising performance in real-world video super-resolution (VSR). However, the dozens of sampling steps they require, make inference extremely slow. Sampling acceleration techniques, particularly single-step, provide a potential solution. Nonetheless, achieving one step in VSR remains challenging, due to the high training overhead on video data and stringent fidelity demands. To tackle the above issues, we propose DOVE, an efficient one-step diffusion model for real-world VSR. DOVE is obtained by fine-tuning a pretrained video diffusion model (i.e., CogVideoX). To effectively train DOVE, we introduce the latent-pixel training strategy. The strategy employs a two-stage scheme to gradually adapt the model to the video super-resolution task. Meanwhile, we design a video processing pipeline to construct a high-quality dataset tailored for VSR, termed HQ-VSR. Fine-tuning on this dataset further enhances the restoration capability of DOVE. Extensive experiments show that DOVE exhibits comparable or superior performance to multi-step diffusion-based VSR methods. It also offers outstanding inference efficiency, achieving up to a 28$\times$ speed-up over existing methods such as MGLD-VSR. Code is available at: https://github.com/zhengchen1999/DOVE.

comment: Accepted to NeurIPS 2025. Code is available at: https://github.com/zhengchen1999/DOVE

♻ ☆ Language-guided Open-world Video Anomaly Detection under Weak Supervision

Zihao Liu, Xiaoyu Wu, Jianqin Wu, Xuxu Wang, Linlin Yang

Video anomaly detection (VAD) aims to detect anomalies that deviate from what is expected. In open-world scenarios, the expected events may change as requirements change. For example, not wearing a mask may be considered abnormal during a flu outbreak but normal otherwise. However, existing methods assume that the definition of anomalies is invariable, and thus are not applicable to the open world. To address this, we propose a novel open-world VAD paradigm with variable definitions, allowing guided detection through user-provided natural language at inference time. This paradigm necessitates establishing a robust mapping from video and textual definition to anomaly scores. Therefore, we propose LaGoVAD (Language-guided Open-world Video Anomaly Detector), a model that dynamically adapts anomaly definitions under weak supervision with two regularization strategies: diversifying the relative durations of anomalies via dynamic video synthesis, and enhancing feature robustness through contrastive learning with negative mining. Training such adaptable models requires diverse anomaly definitions, but existing datasets typically provide labels without semantic descriptions. To bridge this gap, we collect PreVAD (Pre-training Video Anomaly Dataset), the largest and most diverse video anomaly dataset to date, featuring 35,279 annotated videos with multi-level category labels and descriptions that explicitly define anomalies. Zero-shot experiments on seven datasets demonstrate LaGoVAD's SOTA performance. Our dataset and code will be released at https://github.com/Kamino666/LaGoVAD-PreVAD.

♻ ☆ Buffer layers for Test-Time Adaptation NeurIPS 2025

Hyeongyu Kim, Geonhui Han, Dosik Hwang

In recent advancements in Test Time Adaptation (TTA), most existing methodologies focus on updating normalization layers to adapt to the test domain. However, the reliance on normalization-based adaptation presents key challenges. First, normalization layers such as Batch Normalization (BN) are highly sensitive to small batch sizes, leading to unstable and inaccurate statistics. Moreover, normalization-based adaptation is inherently constrained by the structure of the pre-trained model, as it relies on training-time statistics that may not generalize well to unseen domains. These issues limit the effectiveness of normalization-based TTA approaches, especially under significant domain shift. In this paper, we introduce a novel paradigm based on the concept of a Buffer layer, which addresses the fundamental limitations of normalization layer updates. Unlike existing methods that modify the core parameters of the model, our approach preserves the integrity of the pre-trained backbone, inherently mitigating the risk of catastrophic forgetting during online adaptation. Through comprehensive experimentation, we demonstrate that our approach not only outperforms traditional methods in mitigating domain shift and enhancing model robustness, but also exhibits strong resilience to forgetting. Furthermore, our Buffer layer is modular and can be seamlessly integrated into nearly all existing TTA frameworks, resulting in consistent performance improvements across various architectures. These findings validate the effectiveness and versatility of the proposed solution in real-world domain adaptation scenarios. The code is available at https://github.com/hyeongyu-kim/Buffer_TTA.

comment: Accepted at NeurIPS 2025

♻ ☆ Towards Predicting Any Human Trajectory In Context NeurIPS 2025

Ryo Fujii, Hideo Saito, Ryo Hachiuma

Predicting accurate future trajectories of pedestrians is essential for autonomous systems but remains a challenging task due to the need for adaptability in different environments and domains. A common approach involves collecting scenario-specific data and performing fine-tuning via backpropagation. However, the need to fine-tune for each new scenario is often impractical for deployment on edge devices. To address this challenge, we introduce \paper, an In-Context Learning (ICL) framework for pedestrian trajectory prediction that enables adaptation without fine-tuning on the scenario-specific data at inference time without requiring weight updates. We propose a spatio-temporal similarity-based example selection (STES) method that selects relevant examples from previously observed trajectories within the same scene by identifying similar motion patterns at corresponding locations. To further refine this selection, we introduce prediction-guided example selection (PG-ES), which selects examples based on both the past trajectory and the predicted future trajectory, rather than relying solely on the past trajectory. This approach allows the model to account for long-term dynamics when selecting examples. Finally, instead of relying on small real-world datasets with limited scenario diversity, we train our model on a large-scale synthetic dataset to enhance its prediction ability by leveraging in-context examples. Extensive experiments demonstrate that TrajICL achieves remarkable adaptation across both in-domain and cross-domain scenarios, outperforming even fine-tuned approaches across multiple public benchmarks. Project Page: https://fujiry0.github.io/TrajICL-project-page/.

comment: NeurIPS 2025

♻ ☆ SPLite Hand: Sparsity-Aware Lightweight 3D Hand Pose Estimation

Yeh Keng Hao, Hsu Tzu Wei, Sun Min

With the increasing ubiquity of AR/VR devices, the deployment of deep learning models on edge devices has become a critical challenge. These devices require real-time inference, low power consumption, and minimal latency. Many framework designers face the conundrum of balancing efficiency and performance. We design a light framework that adopts an encoder-decoder architecture and introduces several key contributions aimed at improving both efficiency and accuracy. We apply sparse convolution on a ResNet-18 backbone to exploit the inherent sparsity in hand pose images, achieving a 42% end-to-end efficiency improvement. Moreover, we propose our SPLite decoder. This new architecture significantly boosts the decoding process's frame rate by 3.1x on the Raspberry Pi 5, while maintaining accuracy on par. To further optimize performance, we apply quantization-aware training, reducing memory usage while preserving accuracy (PA-MPJPE increases only marginally from 9.0 mm to 9.1 mm on FreiHAND). Overall, our system achieves a 2.98x speed-up on a Raspberry Pi 5 CPU (BCM2712 quad-core Arm A76 processor). Our method is also evaluated on compound benchmark datasets, demonstrating comparable accuracy to state-of-the-art approaches while significantly enhancing computational efficiency.

comment: Accepted to AICCC 2025

♻ ☆ From One to More: Contextual Part Latents for 3D Generation

Shaocong Dong, Lihe Ding, Xiao Chen, Yaokun Li, Yuxin Wang, Yucheng Wang, Qi Wang, Jaehyeok Kim, Chenjian Gao, Zhanpeng Huang, Zibin Wang, Tianfan Xue, Dan Xu

Recent advances in 3D generation have transitioned from multi-view 2D rendering approaches to 3D-native latent diffusion frameworks that exploit geometric priors in ground truth data. Despite progress, three key limitations persist: (1) Single-latent representations fail to capture complex multi-part geometries, causing detail degradation; (2) Holistic latent coding neglects part independence and interrelationships critical for compositional design; (3) Global conditioning mechanisms lack fine-grained controllability. Inspired by human 3D design workflows, we propose CoPart - a part-aware diffusion framework that decomposes 3D objects into contextual part latents for coherent multi-part generation. This paradigm offers three advantages: i) Reduces encoding complexity through part decomposition; ii) Enables explicit part relationship modeling; iii) Supports part-level conditioning. We further develop a mutual guidance strategy to fine-tune pre-trained diffusion models for joint part latent denoising, ensuring both geometric coherence and foundation model priors. To enable large-scale training, we construct Partverse - a novel 3D part dataset derived from Objaverse through automated mesh segmentation and human-verified annotations. Extensive experiments demonstrate CoPart's superior capabilities in part-level editing, articulated object generation, and scene composition with unprecedented controllability.

comment: Project page: https://copart3d.github.io/

♻ ☆ Cycle Diffusion Model for Counterfactual Image Generation

Fangrui Huang, Alan Wang, Binxu Li, Bailey Trang, Ridvan Yesiloglu, Tianyu Hua, Wei Peng, Ehsan Adeli

Deep generative models have demonstrated remarkable success in medical image synthesis. However, ensuring conditioning faithfulness and high-quality synthetic images for direct or counterfactual generation remains a challenge. In this work, we introduce a cycle training framework to fine-tune diffusion models for improved conditioning adherence and enhanced synthetic image realism. Our approach, Cycle Diffusion Model (CDM), enforces consistency between generated and original images by incorporating cycle constraints, enabling more reliable direct and counterfactual generation. Experiments on a combined 3D brain MRI dataset (from ABCD, HCP aging & young adults, ADNI, and PPMI) show that our method improves conditioning accuracy and enhances image quality as measured by FID and SSIM. The results suggest that the cycle strategy used in CDM can be an effective method for refining diffusion-based medical image generation, with applications in data augmentation, counterfactual, and disease progression modeling.

♻ ☆ Empowering Agentic Video Analytics Systems with Video Language Models

Yuxuan Yan, Shiqi Jiang, Ting Cao, Yifan Yang, Qianqian Yang, Yuanchao Shu, Yuqing Yang, Lili Qiu

AI-driven video analytics has become increasingly important across diverse domains. However, existing systems are often constrained to specific, predefined tasks, limiting their adaptability in open-ended analytical scenarios. The recent emergence of Vision Language Models (VLMs) as transformative technologies offers significant potential for enabling open-ended video understanding, reasoning, and analytics. Nevertheless, their limited context windows present challenges when processing ultra-long video content, which is prevalent in real-world applications. To address this, we introduce AVA, a VLM-powered system designed for open-ended, advanced video analytics. AVA incorporates two key innovations: (1) the near real-time construction of Event Knowledge Graphs (EKGs) for efficient indexing of long or continuous video streams, and (2) an agentic retrieval-generation mechanism that leverages EKGs to handle complex and diverse queries. Comprehensive evaluations on public benchmarks, LVBench and VideoMME-Long, demonstrate that AVA achieves state-of-the-art performance, attaining 62.3% and 64.1% accuracy, respectively-significantly surpassing existing VLM and video Retrieval-Augmented Generation (RAG) systems. Furthermore, to evaluate video analytics in ultra-long and open-world video scenarios, we introduce a new benchmark, AVA-100. This benchmark comprises 8 videos, each exceeding 10 hours in duration, along with 120 manually annotated, diverse, and complex question-answer pairs. On AVA-100, AVA achieves top-tier performance with an accuracy of 75.8%. The source code of AVA is available at https://github.com/I-ESC/Project-Ava. The AVA-100 benchmark can be accessed at https://huggingface.co/datasets/iesc/Ava-100.

comment: Accepted to NDSI 2026, 19pages, 12 figures, complementary evaluations and appendix

♻ ☆ Unleashing Diffusion Transformers for Visual Correspondence by Modulating Massive Activations NeurIPS 2025

Chaofan Gan, Yuanpeng Tu, Xi Chen, Tieyuan Chen, Yuxi Li, Mehrtash Harandi, Weiyao Lin

Pre-trained stable diffusion models (SD) have shown great advances in visual correspondence. In this paper, we investigate the capabilities of Diffusion Transformers (DiTs) for accurate dense correspondence. Distinct from SD, DiTs exhibit a critical phenomenon in which very few feature activations exhibit significantly larger values than others, known as \textit{massive activations}, leading to uninformative representations and significant performance degradation for DiTs. The massive activations consistently concentrate at very few fixed dimensions across all image patch tokens, holding little local information. We trace these dimension-concentrated massive activations and find that such concentration can be effectively localized by the zero-initialized Adaptive Layer Norm (AdaLN-zero). Building on these findings, we propose Diffusion Transformer Feature (DiTF), a training-free framework designed to extract semantic-discriminative features from DiTs. Specifically, DiTF employs AdaLN to adaptively localize and normalize massive activations with channel-wise modulation. In addition, we develop a channel discard strategy to further eliminate the negative impacts from massive activations. Experimental results demonstrate that our DiTF outperforms both DINO and SD-based models and establishes a new state-of-the-art performance for DiTs in different visual correspondence tasks (\eg, with +9.4\% on Spair-71k and +4.4\% on AP-10K-C.S.).

comment: NeurIPS 2025

♻ ☆ MMEdge: Accelerating On-device Multimodal Inference via Pipelined Sensing and Encoding

Runxi Huang, Mingxuan Yu, Mingyu Tsoi, Xiaomin Ouyang

Real-time multimodal inference on resource-constrained edge devices is essential for applications such as autonomous driving, human-computer interaction, and mobile health. However, prior work often overlooks the tight coupling between sensing dynamics and model execution, as well as the complex inter-modality dependencies. In this paper, we propose MMEdge, an new on-device multi-modal inference framework based on pipelined sensing and encoding. Instead of waiting for complete sensor inputs, MMEdge decomposes the entire inference process into a sequence of fine-grained sensing and encoding units, allowing computation to proceed incrementally as data arrive. MMEdge also introduces a lightweight but effective temporal aggregation module that captures rich temporal dynamics across different pipelined units to maintain accuracy performance. Such pipelined design also opens up opportunities for fine-grained cross-modal optimization and early decision-making during inference. To further enhance system performance under resource variability and input data complexity, MMEdge incorporates an adaptive multimodal configuration optimizer that dynamically selects optimal sensing and model configurations for each modality under latency constraints, and a cross-modal speculative skipping mechanism that bypasses future units of slower modalities when early predictions reach sufficient confidence. We evaluate MMEdge using two public multimodal datasets and deploy it on a real-world unmanned aerial vehicle (UAV)-based multimodal testbed. The results show that MMEdge significantly reduces end-to-end latency while maintaining high task accuracy across various system and data dynamics.

comment: Code available at: https://github.com/HKUST-MINSys-Lab/MMEdge. Accepted by SenSys 2026

♻ ☆ Boosting Generative Adversarial Transferability with Self-supervised Vision Transformer Features ICCV 2025

Shangbo Wu, Yu-an Tan, Ruinan Ma, Wencong Ma, Dehua Zhu, Yuanzhang Li

The ability of deep neural networks (DNNs) come from extracting and interpreting features from the data provided. By exploiting intermediate features in DNNs instead of relying on hard labels, we craft adversarial perturbation that generalize more effectively, boosting black-box transferability. These features ubiquitously come from supervised learning in previous work. Inspired by the exceptional synergy between self-supervised learning and the Transformer architecture, this paper explores whether exploiting self-supervised Vision Transformer (ViT) representations can improve adversarial transferability. We present dSVA -- a generative dual self-supervised ViT features attack, that exploits both global structural features from contrastive learning (CL) and local textural features from masked image modeling (MIM), the self-supervised learning paradigm duo for ViTs. We design a novel generative training framework that incorporates a generator to create black-box adversarial examples, and strategies to train the generator by exploiting joint features and the attention mechanism of self-supervised ViTs. Our findings show that CL and MIM enable ViTs to attend to distinct feature tendencies, which, when exploited in tandem, boast great adversarial generalizability. By disrupting dual deep features distilled by self-supervised ViTs, we are rewarded with remarkable black-box transferability to models of various architectures that outperform state-of-the-arts. Code available at https://github.com/spencerwooo/dSVA.

comment: 14 pages, 9 figures, accepted at ICCV 2025

♻ ☆ Signal-SGN: A Spiking Graph Convolutional Network for Skeletal Action Recognition via Learning Temporal-Frequency Dynamics

Naichuan Zheng, Yuchen Du, Hailun Xia, Zeyu Liang

For multimodal skeleton-based action recognition, Graph Convolutional Networks (GCNs) are effective models. Still, their reliance on floating-point computations leads to high energy consumption, limiting their applicability in battery-powered devices. While energy-efficient, Spiking Neural Networks (SNNs) struggle to model skeleton dynamics, leading to suboptimal solutions. We propose Signal-SGN (Spiking Graph Convolutional Network), which utilizes the temporal dimension of skeleton sequences as the spike time steps and represents features as multi-dimensional discrete stochastic signals for temporal-frequency domain feature extraction. It combines the 1D Spiking Graph Convolution (1D-SGC) module and the Frequency Spiking Convolution (FSC) module to extract features from the skeleton represented as spiking form. Additionally, the Multi-Scale Wavelet Transform Feature Fusion (MWTF) module is proposed to extract dynamic spiking features and capture frequency-specific characteristics, enhancing classification performance. Experiments across three large-scale datasets reveal Signal-SGN exceeding state-of-the-art SNN-based methods in accuracy and computational efficiency while attaining comparable performance with GCN methods and significantly reducing theoretical energy consumption.

♻ ☆ DiffVLA++: Bridging Cognitive Reasoning and End-to-End Driving through Metric-Guided Alignment

Yu Gao, Anqing Jiang, Yiru Wang, Wang Jijun, Hao Jiang, Zhigang Sun, Heng Yuwen, Wang Shuo, Hao Zhao, Sun Hao

Conventional end-to-end (E2E) driving models are effective at generating physically plausible trajectories, but often fail to generalize to long-tail scenarios due to the lack of essential world knowledge to understand and reason about surrounding environments. In contrast, Vision-Language-Action (VLA) models leverage world knowledge to handle challenging cases, but their limited 3D reasoning capability can lead to physically infeasible actions. In this work we introduce DiffVLA++, an enhanced autonomous driving framework that explicitly bridges cognitive reasoning and E2E planning through metric-guided alignment. First, we build a VLA module directly generating semantically grounded driving trajectories. Second, we design an E2E module with a dense trajectory vocabulary that ensures physical feasibility. Third, and most critically, we introduce a metric-guided trajectory scorer that guides and aligns the outputs of the VLA and E2E modules, thereby integrating their complementary strengths. The experiment on the ICCV 2025 Autonomous Grand Challenge leaderboard shows that DiffVLA++ achieves EPDMS of 49.12.

♻ ☆ ChartMuseum: Testing Visual Reasoning Capabilities of Large Vision-Language Models NeurIPS 2025

Liyan Tang, Grace Kim, Xinyu Zhao, Thom Lake, Wenxuan Ding, Fangcong Yin, Prasann Singhal, Manya Wadhwa, Zeyu Leo Liu, Zayne Sprague, Ramya Namuduri, Bodun Hu, Juan Diego Rodriguez, Puyuan Peng, Greg Durrett

Chart understanding presents a unique challenge for large vision-language models (LVLMs), as it requires the integration of sophisticated textual and visual reasoning capabilities. However, current LVLMs exhibit a notable imbalance between these skills, falling short on visual reasoning that is difficult to perform in text. We conduct a case study using a synthetic dataset solvable only through visual reasoning and show that model performance degrades significantly with increasing visual complexity, while human performance remains robust. We then introduce ChartMuseum, a new Chart Question Answering (QA) benchmark containing 1,162 expert-annotated questions spanning multiple reasoning types, curated from real-world charts across 184 sources, specifically built to evaluate complex visual and textual reasoning. Unlike prior chart understanding benchmarks -- where frontier models perform similarly and near saturation -- our benchmark exposes a substantial gap between model and human performance, while effectively differentiating model capabilities: although humans achieve 93% accuracy, the best-performing model Gemini-2.5-Pro attains only 63.0%, and the leading open-source LVLM Qwen2.5-VL-72B-Instruct achieves only 38.5%. Moreover, on questions requiring primarily visual reasoning, all models experience a 35%-55% performance drop from text-reasoning-heavy question performance. Lastly, our qualitative error analysis reveals specific categories of visual reasoning that are challenging for current LVLMs.

comment: NeurIPS 2025 Datasets & Benchmarks

♻ ☆ Neighborhood Feature Pooling for Remote Sensing Image Classification

Fahimeh Orvati Nia, Amirmohammad Mohammadi, Salim Al Kharsa, Pragati Naikare, Zigfried Hampel-Arias, Joshua Peeples

In this work, we propose neighborhood feature pooling (NFP) as a novel texture feature extraction method for remote sensing image classification. The NFP layer captures relationships between neighboring inputs and efficiently aggregates local similarities across feature dimensions. Implemented using convolutional layers, NFP can be seamlessly integrated into any network. Results comparing the baseline models and the NFP method indicate that NFP consistently improves performance across diverse datasets and architectures while maintaining minimal parameter overhead.

comment: 9 pages, 5 figures

♻ ☆ GCVAMD: A Modified CausalVAE Model for Causal Age-related Macular Degeneration Risk Factor Detection and Prediction

Daeyoung Kim

Age Related Macular Degeneration(AMD) has been one of the most leading causes of permanent vision impairment in ophthalmology. Though treatments, such as anti VEGF drugs or photodynamic therapies, were developed to slow down the degenerative process of AMD, there is still no specific cure to reverse vision loss caused by AMD. Thus, for AMD, detecting existence of risk factors of AMD or AMD itself within the patient retina in early stages is a crucial task to reduce the possibility of vision impairment. Apart from traditional approaches, deep learning based methods, especially attention mechanism based CNNs and GradCAM based XAI analysis on OCT scans, exhibited successful performance in distinguishing AMD retina from normal retinas, making it possible to use AI driven models to aid medical diagnosis and analysis by ophthalmologists regarding AMD. However, though having significant success, previous works mostly focused on prediction performance itself, not pathologies or underlying causal mechanisms of AMD, which can prohibit intervention analysis on specific factors or even lead to less reliable decisions. Thus, this paper introduces a novel causal AMD analysis model: GCVAMD, which incorporates a modified CausalVAE approach that can extract latent causal factors from only raw OCT images. By considering causality in AMD detection, GCVAMD enables causal inference such as treatment simulation or intervention analysis regarding major risk factors: drusen and neovascularization, while returning informative latent causal features that can enhance downstream tasks. Results show that through GCVAMD, drusen status and neovascularization status can be identified with AMD causal mechanisms in GCVAMD latent spaces, which can in turn be used for various tasks from AMD detection(classification) to intervention analysis.

♻ ☆ GameFactory: Creating New Games with Generative Interactive Videos ICCV 2025

Jiwen Yu, Yiran Qin, Xintao Wang, Pengfei Wan, Di Zhang, Xihui Liu

Generative videos have the potential to revolutionize game development by autonomously creating new content. In this paper, we present GameFactory, a framework for action-controlled scene-generalizable game video generation. We first address the fundamental challenge of action controllability by introducing GF-Minecraft, an action-annotated game video dataset without human bias, and developing an action control module that enables precise control over both keyboard and mouse inputs. We further extend to support autoregressive generation for unlimited-length interactive videos. More importantly, GameFactory tackles the critical challenge of scene-generalizable action control, which most existing methods fail to address. To enable the creation of entirely new and diverse games beyond fixed styles and scenes, we leverage the open-domain generative priors from pre-trained video diffusion models. To bridge the domain gap between open-domain priors and small-scale game datasets, we propose a multi-phase training strategy with a domain adapter that decouples game style learning from action control. This decoupling ensures that action control learning is no longer bound to specific game styles, thereby achieving scene-generalizable action control. Experimental results demonstrate that GameFactory effectively generates open-domain action-controllable game videos, representing a significant step forward in AI-driven game generation.

comment: ICCV 2025 Highlight, Project Page: https://yujiwen.github.io/gamefactory

♻ ☆ Reasoning Visual Language Model for Chest X-Ray Analysis

Andriy Myronenko, Dong Yang, Baris Turkbey, Mariam Aboian, Sena Azamat, Esra Akcicek, Hongxu Yin, Pavlo Molchanov, Marc Edgar, Yufan He, Pengfei Guo, Yucheng Tang, Daguang Xu

Vision-language models (VLMs) have shown strong promise for medical image analysis, but most remain opaque, offering predictions without the transparent, stepwise reasoning clinicians rely on. We present a framework that brings chain-of-thought (CoT) reasoning to chest X-ray interpretation. Inspired by reasoning-first training paradigms, our approach is designed to learn how experts reason, not just what they conclude, by aligning intermediate steps with observable image evidence and radiology workflow. Beyond accuracy, the explicit reasoning traces support clinical auditability: they reveal why a conclusion was reached, which alternatives were considered, and where uncertainty remains, enabling quality assurance, error analysis, and safer human-AI collaboration. Our model couples high-fidelity visual encoding with a two-stage training recipe: a reasoning-style supervised fine-tuning (SFT) followed by reinforcement learning (RL) that uses verifiable rewards over a list of X-ray abnormalities. The model outputs reasoning that mirrors radiologists systematic thought process, uncertainty, and differential diagnosis. In out-of-distribution evaluation, the approach achieves competitive multi-label classification while improving interpretability. In a reader study with expert radiologists, full reasoning traces increased confidence, supported error auditing, and reduced time to finalize reports. We release code and the model NV-Reason-CXR-3B to support community progress toward trustworthy, explainable AI in chest radiography and other medical imaging tasks where reasoning quality is as critical as prediction quality.

comment: NV-Reason-CXR-3B

☆ ProfOlaf: Semi-Automated Tool for Systematic Literature Reviews

Martim Afonso, Nuno Saavedra, Bruno Lourenço, Alexandra Mendes, João Ferreira

Systematic reviews and mapping studies are critical for synthesizing research, identifying gaps, and guiding future work, but they are often labor-intensive and time-consuming. Existing tools provide partial support for specific steps, leaving much of the process manual and error-prone. We present ProfOlaf, a semi-automated tool designed to streamline systematic reviews while maintaining methodological rigor. ProfOlaf supports iterative snowballing for article collection with human-in-the-loop filtering and uses large language models to assist in analyzing articles, extracting key topics, and answering queries about the content of papers. By combining automation with guided manual effort, ProfOlaf enhances the efficiency, quality, and reproducibility of systematic reviews across research fields. A video describing and demonstrating ProfOlaf is available at: https://youtu.be/4noUXfcmxsE

comment: 4 pages, 1 Figure, 2 tables

☆ AdSum: Two-stream Audio-visual Summarization for Automated Video Advertisement Clipping

Wen Xie, Yanjun Zhu, Gijs Overgoor, Yakov Bart, Agata Lapedriza Garcia, Sarah Ostadabbas

Advertisers commonly need multiple versions of the same advertisement (ad) at varying durations for a single campaign. The traditional approach involves manually selecting and re-editing shots from longer video ads to create shorter versions, which is labor-intensive and time-consuming. In this paper, we introduce a framework for automated video ad clipping using video summarization techniques. We are the first to frame video clipping as a shot selection problem, tailored specifically for advertising. Unlike existing general video summarization methods that primarily focus on visual content, our approach emphasizes the critical role of audio in advertising. To achieve this, we develop a two-stream audio-visual fusion model that predicts the importance of video frames, where importance is defined as the likelihood of a frame being selected in the firm-produced short ad. To address the lack of ad-specific datasets, we present AdSum204, a novel dataset comprising 102 pairs of 30-second and 15-second ads from real advertising campaigns. Extensive experiments demonstrate that our model outperforms state-of-the-art methods across various metrics, including Average Precision, Area Under Curve, Spearman, and Kendall.

comment: Accepted at 32nd International Conference on MultiMedia Modeling

☆ WeaveRec: An LLM-Based Cross-Domain Sequential Recommendation Framework with Model Merging

Min Hou, Xin Liu, Le Wu, Chenyi He, Hao Liu, Zhi Li, Xin Li, Si Wei

Cross-Domain Sequential Recommendation (CDSR) seeks to improve user preference modeling by transferring knowledge from multiple domains. Despite the progress made in CDSR, most existing methods rely on overlapping users or items to establish cross-domain correlations-a requirement that rarely holds in real-world settings. The advent of large language models (LLM) and model-merging techniques appears to overcome this limitation by unifying multi-domain data without explicit overlaps. Yet, our empirical study shows that naively training an LLM on combined domains-or simply merging several domain-specific LLMs-often degrades performance relative to a model trained solely on the target domain. To address these challenges, we first experimentally investigate the cause of suboptimal performance in LLM-based cross-domain recommendation and model merging. Building on these insights, we introduce WeaveRec, which cross-trains multiple LoRA modules with source and target domain data in a weaving fashion, and fuses them via model merging. WeaveRec can be extended to multi-source domain scenarios and notably does not introduce additional inference-time cost in terms of latency or memory. Furthermore, we provide a theoretical guarantee that WeaveRec can reduce the upper bound of the expected error in the target domain. Extensive experiments on single-source, multi-source, and cross-platform cross-domain recommendation scenarios validate that WeaveRec effectively mitigates performance degradation and consistently outperforms baseline approaches in real-world recommendation tasks.

☆ Inside CORE-KG: Evaluating Structured Prompting and Coreference Resolution for Knowledge Graphs ICDM 2025

Dipak Meher, Carlotta Domeniconi

Human smuggling networks are increasingly adaptive and difficult to analyze. Legal case documents offer critical insights but are often unstructured, lexically dense, and filled with ambiguous or shifting references, which pose significant challenges for automated knowledge graph (KG) construction. While recent LLM-based approaches improve over static templates, they still generate noisy, fragmented graphs with duplicate nodes due to the absence of guided extraction and coreference resolution. The recently proposed CORE-KG framework addresses these limitations by integrating a type-aware coreference module and domain-guided structured prompts, significantly reducing node duplication and legal noise. In this work, we present a systematic ablation study of CORE-KG to quantify the individual contributions of its two key components. Our results show that removing coreference resolution results in a 28.32% increase in node duplication and a 4.32% increase in noisy nodes, while removing structured prompts leads to a 4.34% increase in node duplication and a 73.33% increase in noisy nodes. These findings offer empirical insights for designing robust LLM-based pipelines for extracting structured representations from complex legal texts.

comment: ICDM 2025 Workshop

☆ LINK-KG: LLM-Driven Coreference-Resolved Knowledge Graphs for Human Smuggling Networks

Dipak Meher, Carlotta Domeniconi, Guadalupe Correa-Cabrera

Human smuggling networks are complex and constantly evolving, making them difficult to analyze comprehensively. Legal case documents offer rich factual and procedural insights into these networks but are often long, unstructured, and filled with ambiguous or shifting references, posing significant challenges for automated knowledge graph (KG) construction. Existing methods either overlook coreference resolution or fail to scale beyond short text spans, leading to fragmented graphs and inconsistent entity linking. We propose LINK-KG, a modular framework that integrates a three-stage, LLM-guided coreference resolution pipeline with downstream KG extraction. At the core of our approach is a type-specific Prompt Cache, which consistently tracks and resolves references across document chunks, enabling clean and disambiguated narratives for structured knowledge graph construction from both short and long legal texts. LINK-KG reduces average node duplication by 45.21% and noisy nodes by 32.22% compared to baseline methods, resulting in cleaner and more coherent graph structures. These improvements establish LINK-KG as a strong foundation for analyzing complex criminal networks.

comment: Accepted in ICKG 2025 Conference, 8 Pages, 2 Figures

☆ Vectorized Context-Aware Embeddings for GAT-Based Collaborative Filtering

Danial Ebrat, Sepideh Ahmadian, Luis Rueda

Recommender systems often struggle with data sparsity and cold-start scenarios, limiting their ability to provide accurate suggestions for new or infrequent users. This paper presents a Graph Attention Network (GAT) based Collaborative Filtering (CF) framework enhanced with Large Language Model (LLM) driven context aware embeddings. Specifically, we generate concise textual user profiles and unify item metadata (titles, genres, overviews) into rich textual embeddings, injecting these as initial node features in a bipartite user item graph. To further optimize ranking performance, we introduce a hybrid loss function that combines Bayesian Personalized Ranking (BPR) with a cosine similarity term and robust negative sampling, ensuring explicit negative feedback is distinguished from unobserved data. Experiments on the MovieLens 100k and 1M datasets show consistent improvements over state-of-the-art baselines in Precision, NDCG, and MAP while demonstrating robustness for users with limited interaction history. Ablation studies confirm the critical role of LLM-augmented embeddings and the cosine similarity term in capturing nuanced semantic relationships. Our approach effectively mitigates sparsity and cold-start limitations by integrating LLM-derived contextual understanding into graph-based architectures. Future directions include balancing recommendation accuracy with coverage and diversity, and introducing fairness-aware constraints and interpretability features to enhance system performance further.

☆ Barlow Twins for Sequential Recommendation

Ivan Razvorotnev, Marina Munkhoeva, Evgeny Frolov

Sequential recommendation models must navigate sparse interaction data popularity bias and conflicting objectives like accuracy versus diversity While recent contrastive selfsupervised learning SSL methods offer improved accuracy they come with tradeoffs large batch requirements reliance on handcrafted augmentations and negative sampling that can reinforce popularity bias In this paper we introduce BT-SR a novel noncontrastive SSL framework that integrates the Barlow Twins redundancyreduction principle into a Transformerbased nextitem recommender BTSR learns embeddings that align users with similar shortterm behaviors while preserving longterm distinctionswithout requiring negative sampling or artificial perturbations This structuresensitive alignment allows BT-SR to more effectively recognize emerging user intent and mitigate the influence of noisy historical context Our experiments on five public benchmarks demonstrate that BTSR consistently improves nextitem prediction accuracy and significantly enhances longtail item coverage and recommendation calibration Crucially we show that a single hyperparameter can control the accuracydiversity tradeoff enabling practitioners to adapt recommendations to specific application needs

☆ GraphCompliance: Aligning Policy and Context Graphs for LLM-Based Regulatory Compliance

Jiseong Chung, Ronny Ko, Wonchul Yoo, Makoto Onizuka, Sungmok Kim, Tae-Wan Kim, Won-Yong Shin

Compliance at web scale poses practical challenges: each request may require a regulatory assessment. Regulatory texts (e.g., the General Data Protection Regulation, GDPR) are cross-referential and normative, while runtime contexts are expressed in unstructured natural language. This setting motivates us to align semantic information in unstructured text with the structured, normative elements of regulations. To this end, we introduce GraphCompliance, a framework that represents regulatory texts as a Policy Graph and runtime contexts as a Context Graph, and aligns them. In this formulation, the policy graph encodes normative structure and cross-references, whereas the context graph formalizes events as subject-action-object (SAO) and entity-relation triples. This alignment anchors the reasoning of a judge large language model (LLM) in structured information and helps reduce the burden of regulatory interpretation and event parsing, enabling a focus on the core reasoning step. In experiments on 300 GDPR-derived real-world scenarios spanning five evaluation tasks, GraphCompliance yields 4.1-7.2 percentage points (pp) higher micro-F1 than LLM-only and RAG baselines, with fewer under- and over-predictions, resulting in higher recall and lower false positive rates. Ablation studies indicate contributions from each graph component, suggesting that structured representations and a judge LLM are complementary for normative reasoning.

comment: Under review at The Web Conference 2026 (Semantics & Knowledge track). Code will be released upon acceptance. This arXiv v1 contains no repository links to preserve double-blind review

☆ DiSE: A diffusion probabilistic model for automatic structure elucidation of organic compounds

Haochen Chen, Qi Huang, Anan Wu, Wenhao Zhang, Jianliang Ye, Jianming Wu, Kai Tan, Xin Lu, Xin Xu

Automatic structure elucidation is essential for self-driving laboratories as it enables the system to achieve truly autonomous. This capability closes the experimental feedback loop, ensuring that machine learning models receive reliable structure information for real-time decision-making and optimization. Herein, we present DiSE, an end-to-end diffusion-based generative model that integrates multiple spectroscopic modalities, including MS, 13C and 1H chemical shifts, HSQC, and COSY, to achieve automated yet accurate structure elucidation of organic compounds. By learning inherent correlations among spectra through data-driven approaches, DiSE achieves superior accuracy, strong generalization across chemically diverse datasets, and robustness to experimental data despite being trained on calculated spectra. DiSE thus represents a significant advance toward fully automated structure elucidation, with broad potential in natural product research, drug discovery, and self-driving laboratories.

☆ ReaKase-8B: Legal Case Retrieval via Knowledge and Reasoning Representations with LLMs

Yanran Tang, Ruihong Qiu, Xue Li, Zi Huang

Legal case retrieval (LCR) is a cornerstone of real-world legal decision making, as it enables practitioners to identify precedents for a given query case. Existing approaches mainly rely on traditional lexical models and pretrained language models to encode the texts of legal cases. Yet there are rich information in the relations among different legal entities as well as the crucial reasoning process that uncovers how legal facts and legal issues can lead to judicial decisions. Such relational reasoning process reflects the distinctive characteristics of each case that can distinguish one from another, mirroring the real-world judicial process. Naturally, incorporating such information into the precise case embedding could further enhance the accuracy of case retrieval. In this paper, a novel ReaKase-8B framework is proposed to leverage extracted legal facts, legal issues, legal relation triplets and legal reasoning for effective legal case retrieval. ReaKase-8B designs an in-context legal case representation learning paradigm with a fine-tuned large language model. Extensive experiments on two benchmark datasets from COLIEE 2022 and COLIEE 2023 demonstrate that our knowledge and reasoning augmented embeddings substantially improve retrieval performance over baseline models, highlighting the potential of integrating legal reasoning into legal case retrieval systems. The code has been released on https://github.com/yanran-tang/ReaKase-8B.

☆ OneTrans: Unified Feature Interaction and Sequence Modeling with One Transformer in Industrial Recommender

Zhaoqi Zhang, Haolei Pei, Jun Guo, Tianyu Wang, Yufei Feng, Hui Sun, Shaowei Liu, Aixin Sun

In recommendation systems, scaling up feature-interaction modules (e.g., Wukong, RankMixer) or user-behavior sequence modules (e.g., LONGER) has achieved notable success. However, these efforts typically proceed on separate tracks, which not only hinders bidirectional information exchange but also prevents unified optimization and scaling. In this paper, we propose OneTrans, a unified Transformer backbone that simultaneously performs user-behavior sequence modeling and feature interaction. OneTrans employs a unified tokenizer to convert both sequential and non-sequential attributes into a single token sequence. The stacked OneTrans blocks share parameters across similar sequential tokens while assigning token-specific parameters to non-sequential tokens. Through causal attention and cross-request KV caching, OneTrans enables precomputation and caching of intermediate representations, significantly reducing computational costs during both training and inference. Experimental results on industrial-scale datasets demonstrate that OneTrans scales efficiently with increasing parameters, consistently outperforms strong baselines, and yields a 5.68% lift in per-user GMV in online A/B tests.

☆ ORBIT -- Open Recommendation Benchmark for Reproducible Research with Hidden Tests NeurIPS 2025

Jingyuan He, Jiongnan Liu, Vishan Vishesh Oberoi, Bolin Wu, Mahima Jagadeesh Patel, Kangrui Mao, Chuning Shi, I-Ta Lee, Arnold Overwijk, Chenyan Xiong

Recommender systems are among the most impactful AI applications, interacting with billions of users every day, guiding them to relevant products, services, or information tailored to their preferences. However, the research and development of recommender systems are hindered by existing datasets that fail to capture realistic user behaviors and inconsistent evaluation settings that lead to ambiguous conclusions. This paper introduces the Open Recommendation Benchmark for Reproducible Research with HIdden Tests (ORBIT), a unified benchmark for consistent and realistic evaluation of recommendation models. ORBIT offers a standardized evaluation framework of public datasets with reproducible splits and transparent settings for its public leaderboard. Additionally, ORBIT introduces a new webpage recommendation task, ClueWeb-Reco, featuring web browsing sequences from 87 million public, high-quality webpages. ClueWeb-Reco is a synthetic dataset derived from real, user-consented, and privacy-guaranteed browsing data. It aligns with modern recommendation scenarios and is reserved as the hidden test part of our leaderboard to challenge recommendation models' generalization ability. ORBIT measures 12 representative recommendation models on its public benchmark and introduces a prompted LLM baseline on the ClueWeb-Reco hidden test. Our benchmark results reflect general improvements of recommender systems on the public datasets, with variable individual performances. The results on the hidden test reveal the limitations of existing approaches in large-scale webpage recommendation and highlight the potential for improvements with LLM integrations. ORBIT benchmark, leaderboard, and codebase are available at https://www.open-reco-bench.ai.

comment: Accepted to NeurIPS 2025 Datasets & Benchmarks track

♻ ☆ Quality Over Quantity? LLM-Based Curation for a Data-Efficient Audio-Video Foundation Model

Ali Vosoughi, Dimitra Emmanouilidou, Hannes Gamper

Integrating audio and visual data for training multimodal foundational models remains a challenge. The Audio-Video Vector Alignment (AVVA) framework addresses this by considering AV scene alignment beyond mere temporal synchronization, and leveraging Large Language Models (LLMs) for data curation. AVVA implements a scoring mechanism for selecting aligned training data segments. It integrates Whisper, a speech-based foundation model, for audio and DINOv2 for video analysis in a dual-encoder structure with contrastive learning on AV pairs. Evaluations on AudioCaps, VALOR, and VGGSound demonstrate the effectiveness of the proposed model architecture and data curation approach. AVVA achieves a significant improvement in top-k accuracies for video-to-audio retrieval on all datasets compared to DenseAV, while using only 192 hrs of curated training data. Furthermore, an ablation study indicates that the data curation process effectively trades data quality for data quantity, yielding increases in top-k retrieval accuracies on AudioCaps, VALOR, and VGGSound, compared to training on the full spectrum of uncurated data.

comment: 5 pages, 5 figures, 2 tables. Accepted at EUSIPCO 2025

♻ ☆ Unveiling Unicode's Unseen Underpinnings in Undermining Authorship Attribution

Robert Dilworth

When using a public communication channel -- whether formal or informal, such as commenting or posting on social media -- end users have no expectation of privacy: they compose a message and broadcast it for the world to see. Even if an end user takes utmost precautions to anonymize their online presence -- using an alias or pseudonym; masking their IP address; spoofing their geolocation; concealing their operating system and user agent; deploying encryption; registering with a disposable phone number or email; disabling non-essential settings; revoking permissions; and blocking cookies and fingerprinting -- one obvious element still lingers: the message itself. Assuming they avoid lapses in judgment or accidental self-exposure, there should be little evidence to validate their actual identity, right? Wrong. The content of their message -- necessarily open for public consumption -- exposes an attack vector: stylometric analysis, or author profiling. In this paper, we dissect the technique of stylometry, discuss an antithetical counter-strategy in adversarial stylometry, and devise enhancements through Unicode steganography.

comment: 33 pages, 7 figures, 3 tables

♻ ☆ Unstructured Evidence Attribution for Long Context Query Focused Summarization EMNLP 2025

Dustin Wright, Zain Muhammad Mujahid, Lu Wang, Isabelle Augenstein, David Jurgens

Large language models (LLMs) are capable of generating coherent summaries from very long contexts given a user query, and extracting and citing evidence spans helps improve the trustworthiness of these summaries. Whereas previous work has focused on evidence citation with fixed levels of granularity (e.g. sentence, paragraph, document, etc.), we propose to extract unstructured (i.e., spans of any length) evidence in order to acquire more relevant and consistent evidence than in the fixed granularity case. We show how existing systems struggle to copy and properly cite unstructured evidence, which also tends to be "lost-in-the-middle". To help models perform this task, we create the Summaries with Unstructured Evidence Text dataset (SUnsET), a synthetic dataset generated using a novel pipeline, which can be used as training supervision for unstructured evidence summarization. We demonstrate across 5 LLMs and 4 datasets spanning human written, synthetic, single, and multi-document settings that LLMs adapted with SUnsET generate more relevant and factually consistent evidence with their summaries, extract evidence from more diverse locations in their context, and can generate more relevant and consistent summaries than baselines with no fine-tuning and fixed granularity evidence. We release SUnsET and our generation code to the public.

comment: EMNLP 2025 Main; 29 pages; 24 figures; 8 tables

♻ ☆ Epistemic Diversity and Knowledge Collapse in Large Language Models

Dustin Wright, Sarah Masud, Jared Moore, Srishti Yadav, Maria Antoniak, Chan Young Park, Isabelle Augenstein

Large language models (LLMs) tend to generate lexically, semantically, and stylistically homogenous texts. This poses a risk of knowledge collapse, where homogenous LLMs mediate a shrinking in the range of accessible information over time. Existing works on homogenization are limited by a focus on closed-ended multiple-choice setups or fuzzy semantic features, and do not look at trends across time and cultural contexts. To overcome this, we present a new methodology to measure epistemic diversity, i.e., variation in real-world claims in LLM outputs, which we use to perform a broad empirical study of LLM knowledge collapse. We test 27 LLMs, 155 topics covering 12 countries, and 200 prompt variations sourced from real user chats. For the topics in our study, we show that while newer models tend to generate more diverse claims, nearly all models are less epistemically diverse than a basic web search. We find that model size has a negative impact on epistemic diversity, while retrieval-augmented generation (RAG) has a positive impact, though the improvement from RAG varies by the cultural context. Finally, compared to a traditional knowledge source (Wikipedia), we find that country-specific claims reflect the English language more than the local one, highlighting a gap in epistemic representation

comment: 16 pages; 8 figures, 4 tables; v2 changelog: Fixed the modeling for table 3, random effect is the model version; v3 changelog: Fixed minor formatting issues in tables 2 and 3; v4 changelog: Fixed some typos and model description

♻ ☆ RecCocktail: A Generalizable and Efficient Framework for LLM-Based Recommendation

Min Hou, Chenxi Bai, Le Wu, Hao Liu, Kai Zhang, Weiwen Liu, Richang Hong, Ruiming Tang, Meng Wang

Large Language Models (LLMs) have achieved remarkable success in recent years, owing to their impressive generalization capabilities and rich world knowledge. To capitalize on the potential of using LLMs as recommender systems, mainstream approaches typically focus on two paradigms. The first paradigm designs multi-domain or multi-task instruction data for generalizable recommendation, so as to align LLMs with general recommendation areas and deal with cold-start recommendation. The second paradigm focuses on enhancing domain-specific recommendation tasks, improving performance in warm recommendation scenarios. While most previous works treat these two paradigms separately, we argue that they have complementary advantages, and combining them can yield better results. In this paper, we propose a generalizable and efficient LLM-based recommendation framework RecCocktail. Our approach begins with fine-tuning a "base spirit" LoRA module using domain-general recommendation instruction data to align LLM with recommendation knowledge. Next, given users' behavior of a specific domain, we construct a domain-specific "ingredient" LoRA module. We then provide an entropy-guided adaptive merging method to mix the "base spirit" and the "ingredient" in the weight space. Please note that, RecCocktail combines the advantages of the existing two paradigms without introducing additional time or space overhead during the inference phase. Moreover, RecCocktail is efficient with plug and play, as the "base spirit" LoRA is trained only once, and any domain-specific "ingredient" can be efficiently mixed with only domain-specific fine-tuning. Extensive experiments on multiple datasets under both warm and cold-start recommendation scenarios validate the effectiveness and generality of the proposed RecCocktail.

♻ ☆ On-the-Fly OVD Adaptation with FLAME: Few-shot Localization via Active Marginal-Samples Exploration

Yehonathan Refael, Amit Aides, Aviad Barzilai, George Leifman, Genady Beryozkin, Vered Silverman, Bolous Jaber, Tomer Shekel

Open-vocabulary object detection (OVD) models offer remarkable flexibility by detecting objects from arbitrary text queries. However, their zero-shot performance in specialized domains like Remote Sensing (RS) is often compromised by the inherent ambiguity of natural language, limiting critical downstream applications. For instance, an OVD model may struggle to distinguish between fine-grained classes such as "fishing boat" and "yacht" since their embeddings are similar and often inseparable. This can hamper specific user goals, such as monitoring illegal fishing, by producing irrelevant detections. To address this, we propose a cascaded approach that couples the broad generalization of a large pre-trained OVD model with a lightweight few-shot classifier. Our method first employs the zero-shot model to generate high-recall object proposals. These proposals are then refined for high precision by a compact classifier trained in real-time on only a handful of user-annotated examples - drastically reducing the high costs of RS imagery annotation.The core of our framework is FLAME, a one-step active learning strategy that selects the most informative samples for training. FLAME identifies, on the fly, uncertain marginal candidates near the decision boundary using density estimation, followed by clustering to ensure sample diversity. This efficient sampling technique achieves high accuracy without costly full-model fine-tuning and enables instant adaptation, within less then a minute, which is significantly faster than state-of-the-art alternatives.Our method consistently surpasses state-of-the-art performance on RS benchmarks, establishing a practical and resource-efficient framework for adapting foundation models to specific user needs.

♻ ☆ Model-Document Protocol for AI Search

Hongjin Qian, Zheng Liu

AI search depends on linking large language models (LLMs) with vast external knowledge sources. Yet web pages, PDF files, and other raw documents are not inherently LLM-ready: they are long, noisy, and unstructured. Conventional retrieval methods treat these documents as verbatim text and return raw passages, leaving the burden of fragment assembly and contextual reasoning to the LLM. This gap underscores the need for a new retrieval paradigm that redefines how models interact with documents. We introduce the Model-Document Protocol (MDP), a general framework that formalizes how raw text is bridged to LLMs through consumable knowledge representations. Rather than treating retrieval as passage fetching, MDP defines multiple pathways that transform unstructured documents into task-specific, LLM-ready inputs. These include agentic reasoning, which curates raw evidence into coherent context; memory grounding, which accumulates reusable notes to enrich reasoning; and structured leveraging, which encodes documents into formal representations such as graphs or key-value caches. All three pathways share the same goal: ensuring that what reaches the LLM is not raw fragments but compact, structured knowledge directly consumable for reasoning. As an instantiation, we present MDP-Agent, which realizes the protocol through an agentic process: constructing document-level gist memories for global coverage, performing diffusion-based exploration with vertical exploitation to uncover layered dependencies, and applying map-reduce style synthesis to integrate large-scale evidence into compact yet sufficient context. Experiments on information-seeking benchmarks demonstrate that MDP-Agent outperforms baselines, validating both the soundness of the MDP framework and the effectiveness of its agentic instantiation.

comment: 10 pages

♻ ☆ The RAG Paradox: A Black-Box Attack Exploiting Unintentional Vulnerabilities in Retrieval-Augmented Generation Systems

Chanwoo Choi, Jinsoo Kim, Sukmin Cho, Soyeong Jeong, Buru Chang

With the growing adoption of retrieval-augmented generation (RAG) systems, various attack methods have been proposed to degrade their performance. However, most existing approaches rely on unrealistic assumptions in which external attackers have access to internal components such as the retriever. To address this issue, we introduce a realistic black-box attack based on the RAG paradox, a structural vulnerability arising from the system's effort to enhance trust by revealing both the retrieved documents and their sources to users. This transparency enables attackers to observe which sources are used and how information is phrased, allowing them to craft poisoned documents that are more likely to be retrieved and upload them to the identified sources. Moreover, as RAG systems directly provide retrieved content to users, these documents must not only be retrievable but also appear natural and credible to maintain user confidence in the search results. Unlike prior work that focuses solely on improving document retrievability, our attack method explicitly considers both retrievability and user trust in the retrieved content. Both offline and online experiments demonstrate that our method significantly degrades system performance without internal access, while generating natural-looking poisoned documents.

♻ ☆ Shilling Recommender Systems by Generating Side-feature-aware Fake User Profiles

Yuanrong Wang, Yingpeng Du

Recommender systems (RS) greatly influence users' consumption decisions, making them attractive targets for malicious shilling attacks that inject fake user profiles to manipulate recommendations. Existing shilling methods can generate effective and stealthy fake profiles when training data only contain rating matrix, but they lack comprehensive solutions for scenarios where side features are present and utilized by the recommender. To address this gap, we extend the Leg-UP framework by enhancing the generator architecture to incorporate side features, enabling the generation of side-feature-aware fake user profiles. Experiments on benchmarks show that our method achieves strong attack performance while maintaining stealthiness.

♻ ☆ MMQ-v2: Align, Denoise, and Amplify: Adaptive Behavior Mining for Semantic IDs Learning in Recommendation

Yi Xu, Moyu Zhang, Chaofan Fan, Jinxin Hu, Xiaochen Li, Yu Zhang, Xiaoyi Zeng, Jing Zhang

Industrial recommender systems rely on unique Item Identifiers (ItemIDs). However, this method struggles with scalability and generalization in large, dynamic datasets that have sparse long-tail data. Content-based Semantic IDs (SIDs) address this by sharing knowledge through content quantization. However, by ignoring dynamic behavioral properties, purely content-based SIDs have limited expressive power. Existing methods attempt to incorporate behavioral information but overlook a critical distinction: unlike relatively uniform content features, user-item interactions are highly skewed and diverse, creating a vast information gap in quality and quantity between popular and long-tail items. This oversight leads to two critical limitations: (1) Noise Corruption: Indiscriminate behavior-content alignment allows collaborative noise from long-tail items to corrupt their content representations, leading to the loss of critical multimodal information. (2)Signal Obscurity: The equal-weighting scheme for SIDs fails to reflect the varying importance of different behavioral signals, making it difficult for downstream tasks to distinguish important SIDs from uninformative ones. To tackle these issues, we propose a mixture-of-quantization framework, MMQ-v2, to adaptively Align, Denoise, and Amplify multimodal information from content and behavior modalities for semantic IDs learning. The semantic IDs generated by this framework named ADA-SID. It introduces two innovations: an adaptive behavior-content alignment that is aware of information richness to shield representations from noise, and a dynamic behavioral router to amplify critical signals by applying different weights to SIDs. Extensive experiments on public and large-scale industrial datasets demonstrate ADA-SID's significant superiority in both generative and discriminative recommendation tasks.

♻ ☆ Towards Automated Quality Assurance of Patent Specifications: A Multi-Dimensional LLM Framework

Yuqian Chai, Chaochao Wang, Weilei Wang

Although AI drafting tools have gained prominence in patent writing, the systematic evaluation of AI-generated patent content quality represents a significant research gap. To address this gap, We propose to evaluate patents using regulatory compliance, technical coherence, and figure-reference consistency detection modules, and then generate improvement suggestions via an integration module. The framework is validated on a comprehensive dataset comprising 80 human-authored and 80 AI-generated patents from two patent drafting tools. Evaluation is performed on 10,841 total sentences, 8,924 non-template sentences, and 554 patent figures for the three detection modules respectively, achieving balanced accuracies of 99.74%, 82.12%, and 91.2% against expert annotations. Additional analysis was conducted to examine defect distributions across patent sections, technical domains, and authoring sources. Section-based analysis indicates that figure-text consistency and technical detail precision require particular attention. Mechanical Engineering and Construction show more claim-specification inconsistencies due to complex technical documentation requirements. AI-generated patents show a significant gap compared to human-authored ones. While human-authored patents primarily contain surface-level errors like typos, AI-generated patents exhibit more structural defects in figure-text alignment and cross-references.

♻ ☆ Decoupled Multimodal Fusion for User Interest Modeling in Click-Through Rate Prediction

Alin Fan, Hanqing Li, Sihan Lu, Jingsong Yuan, Jiandong Zhang

Modern industrial recommendation systems improve recommendation performance by integrating multimodal representations from pre-trained models into ID-based Click-Through Rate (CTR) prediction frameworks. However, existing approaches typically adopt modality-centric modeling strategies that process ID-based and multimodal embeddings independently, failing to capture fine-grained interactions between content semantics and behavioral signals. In this paper, we propose Decoupled Multimodal Fusion (DMF), which introduces a modality-enriched modeling strategy to enable fine-grained interactions between ID-based collaborative representations and multimodal representations for user interest modeling. Specifically, we construct target-aware features to bridge the semantic gap across different embedding spaces and leverage them as side information to enhance the effectiveness of user interest modeling. Furthermore, we design an inference-optimized attention mechanism that decouples the computation of target-aware features and ID-based embeddings before the attention layer, thereby alleviating the computational bottleneck introduced by incorporating target-aware features. To achieve comprehensive multimodal integration, DMF combines user interest representations learned under the modality-centric and modality-enriched modeling strategies. Offline experiments on public and industrial datasets demonstrate the effectiveness of DMF. Moreover, DMF has been deployed on the product recommendation system of the international e-commerce platform Lazada, achieving relative improvements of 5.30% in CTCVR and 7.43% in GMV with negligible computational overhead.

♻ ☆ A Task-Centric Perspective on Recommendation Systems

Aixin Sun

Many studies in recommender systems (RecSys) adopt a general problem definition, i.e., to recommend preferred items to users based on past interactions. Such abstraction often lacks the domain-specific nuances necessary for practical deployment. However, models are frequently evaluated using datasets collected from online recommender platforms, which inherently reflect domain or task specificities. In this paper, we analyze RecSys task formulations, emphasizing key components such as input-output structures, temporal dynamics, and candidate item selection. All these factors directly impact offline evaluation. We further examine the complexities of user-item interactions, including decision-making costs, multi-step engagements, and unobservable interactions, which may influence model design. Additionally, we explore the balance between task specificity and model generalizability, highlighting how well-defined task formulations serve as the foundation for robust evaluation and effective solution development. By clarifying task definitions and their implications, this work provides a structured perspective on RecSys research. The goal is to help researchers better navigate the field, particularly in understanding specificities of the RecSys tasks and ensuring fair and meaningful evaluations.

☆ OmniX: From Unified Panoramic Generation and Perception to Graphics-Ready 3D Scenes

Yukun Huang, Jiwen Yu, Yanning Zhou, Jianan Wang, Xintao Wang, Pengfei Wan, Xihui Liu

There are two prevalent ways to constructing 3D scenes: procedural generation and 2D lifting. Among them, panorama-based 2D lifting has emerged as a promising technique, leveraging powerful 2D generative priors to produce immersive, realistic, and diverse 3D environments. In this work, we advance this technique to generate graphics-ready 3D scenes suitable for physically based rendering (PBR), relighting, and simulation. Our key insight is to repurpose 2D generative models for panoramic perception of geometry, textures, and PBR materials. Unlike existing 2D lifting approaches that emphasize appearance generation and ignore the perception of intrinsic properties, we present OmniX, a versatile and unified framework. Based on a lightweight and efficient cross-modal adapter structure, OmniX reuses 2D generative priors for a broad range of panoramic vision tasks, including panoramic perception, generation, and completion. Furthermore, we construct a large-scale synthetic panorama dataset containing high-quality multimodal panoramas from diverse indoor and outdoor scenes. Extensive experiments demonstrate the effectiveness of our model in panoramic visual perception and graphics-ready 3D scene generation, opening new possibilities for immersive and physically realistic virtual world generation.

comment: Project page: https://yukun-huang.github.io/OmniX/

☆ Scaling Image Geo-Localization to Continent Level NeurIPS 2025

Philipp Lindenberger, Paul-Edouard Sarlin, Jan Hosang, Matteo Balice, Marc Pollefeys, Simon Lynen, Eduard Trulls

Determining the precise geographic location of an image at a global scale remains an unsolved challenge. Standard image retrieval techniques are inefficient due to the sheer volume of images (>100M) and fail when coverage is insufficient. Scalable solutions, however, involve a trade-off: global classification typically yields coarse results (10+ kilometers), while cross-view retrieval between ground and aerial imagery suffers from a domain gap and has been primarily studied on smaller regions. This paper introduces a hybrid approach that achieves fine-grained geo-localization across a large geographic expanse the size of a continent. We leverage a proxy classification task during training to learn rich feature representations that implicitly encode precise location information. We combine these learned prototypes with embeddings of aerial imagery to increase robustness to the sparsity of ground-level data. This enables direct, fine-grained retrieval over areas spanning multiple countries. Our extensive evaluation demonstrates that our approach can localize within 200m more than 68\% of queries of a dataset covering a large part of Europe. The code is publicly available at https://scaling-geoloc.github.io.

comment: NeurIPS 2025

☆ Learning Pseudorandom Numbers with Transformers: Permuted Congruential Generators, Curricula, and Interpretability

Tao Tao, Maissam Barkeshli

We study the ability of Transformer models to learn sequences generated by Permuted Congruential Generators (PCGs), a widely used family of pseudo-random number generators (PRNGs). PCGs introduce substantial additional difficulty over linear congruential generators (LCGs) by applying a series of bit-wise shifts, XORs, rotations and truncations to the hidden state. We show that Transformers can nevertheless successfully perform in-context prediction on unseen sequences from diverse PCG variants, in tasks that are beyond published classical attacks. In our experiments we scale moduli up to $2^{22}$ using up to $50$ million model parameters and datasets with up to $5$ billion tokens. Surprisingly, we find even when the output is truncated to a single bit, it can be reliably predicted by the model. When multiple distinct PRNGs are presented together during training, the model can jointly learn them, identifying structures from different permutations. We demonstrate a scaling law with modulus $m$: the number of in-context sequence elements required for near-perfect prediction grows as $\sqrt{m}$. For larger moduli, optimization enters extended stagnation phases; in our experiments, learning moduli $m \geq 2^{20}$ requires incorporating training data from smaller moduli, demonstrating a critical necessity for curriculum learning. Finally, we analyze embedding layers and uncover a novel clustering phenomenon: the model spontaneously groups the integer inputs into bitwise rotationally-invariant clusters, revealing how representations can transfer from smaller to larger moduli.

comment: 10+13 pages, 8+19 figures

☆ Defeating the Training-Inference Mismatch via FP16

Penghui Qi, Zichen Liu, Xiangxin Zhou, Tianyu Pang, Chao Du, Wee Sun Lee, Min Lin

Reinforcement learning (RL) fine-tuning of large language models (LLMs) often suffers from instability due to the numerical mismatch between the training and inference policies. While prior work has attempted to mitigate this issue through algorithmic corrections or engineering alignments, we show that its root cause lies in the floating point precision itself. The widely adopted BF16, despite its large dynamic range, introduces large rounding errors that breaks the consistency between training and inference. In this work, we demonstrate that simply reverting to \textbf{FP16} effectively eliminates this mismatch. The change is simple, fully supported by modern frameworks with only a few lines of code change, and requires no modification to the model architecture or learning algorithm. Our results suggest that using FP16 uniformly yields more stable optimization, faster convergence, and stronger performance across diverse tasks, algorithms and frameworks. We hope these findings motivate a broader reconsideration of precision trade-offs in RL fine-tuning.

☆ Remote Labor Index: Measuring AI Automation of Remote Work

Mantas Mazeika, Alice Gatti, Cristina Menghini, Udari Madhushani Sehwag, Shivam Singhal, Yury Orlovskiy, Steven Basart, Manasi Sharma, Denis Peskoff, Elaine Lau, Jaehyuk Lim, Lachlan Carroll, Alice Blair, Vinaya Sivakumar, Sumana Basu, Brad Kenstler, Yuntao Ma, Julian Michael, Xiaoke Li, Oliver Ingebretsen, Aditya Mehta, Jean Mottola, John Teichmann, Kevin Yu, Zaina Shaik, Adam Khoja, Richard Ren, Jason Hausenloy, Long Phan, Ye Htet, Ankit Aich, Tahseen Rabbani, Vivswan Shah, Andriy Novykov, Felix Binder, Kirill Chugunov, Luis Ramirez, Matias Geralnik, Hernán Mesura, Dean Lee, Ed-Yeremai Hernandez Cardona, Annette Diamond, Summer Yue, Alexandr Wang, Bing Liu, Ernesto Hernandez, Dan Hendrycks

AIs have made rapid progress on research-oriented benchmarks of knowledge and reasoning, but it remains unclear how these gains translate into economic value and automation. To measure this, we introduce the Remote Labor Index (RLI), a broadly multi-sector benchmark comprising real-world, economically valuable projects designed to evaluate end-to-end agent performance in practical settings. AI agents perform near the floor on RLI, with the highest-performing agent achieving an automation rate of 2.5%. These results help ground discussions of AI automation in empirical evidence, setting a common basis for tracking AI impacts and enabling stakeholders to proactively navigate AI-driven labor automation.

comment: Website: https://www.remotelabor.ai

☆ HEIR: Learning Graph-Based Motion Hierarchies

Cheng Zheng, William Koch, Baiang Li, Felix Heide

Hierarchical structures of motion exist across research fields, including computer vision, graphics, and robotics, where complex dynamics typically arise from coordinated interactions among simpler motion components. Existing methods to model such dynamics typically rely on manually-defined or heuristic hierarchies with fixed motion primitives, limiting their generalizability across different tasks. In this work, we propose a general hierarchical motion modeling method that learns structured, interpretable motion relationships directly from data. Our method represents observed motions using graph-based hierarchies, explicitly decomposing global absolute motions into parent-inherited patterns and local motion residuals. We formulate hierarchy inference as a differentiable graph learning problem, where vertices represent elemental motions and directed edges capture learned parent-child dependencies through graph neural networks. We evaluate our hierarchical reconstruction approach on three examples: 1D translational motion, 2D rotational motion, and dynamic 3D scene deformation via Gaussian splatting. Experimental results show that our method reconstructs the intrinsic motion hierarchy in 1D and 2D cases, and produces more realistic and interpretable deformations compared to the baseline on dynamic 3D Gaussian splatting scenes. By providing an adaptable, data-driven hierarchical modeling paradigm, our method offers a formulation applicable to a broad range of motion-centric tasks. Project Page: https://light.princeton.edu/HEIR/

comment: Code link: https://github.com/princeton-computational-imaging/HEIR

☆ A Unified Theory for Causal Inference: Direct Debiased Machine Learning via Bregman-Riesz Regression

Masahiro Kato

This note introduces a unified theory for causal inference that integrates Riesz regression, covariate balancing, density-ratio estimation (DRE), targeted maximum likelihood estimation (TMLE), and the matching estimator in average treatment effect (ATE) estimation. In ATE estimation, the balancing weights and the regression functions of the outcome play important roles, where the balancing weights are referred to as the Riesz representer, bias-correction term, and clever covariates, depending on the context. Riesz regression, covariate balancing, DRE, and the matching estimator are methods for estimating the balancing weights, where Riesz regression is essentially equivalent to DRE in the ATE context, the matching estimator is a special case of DRE, and DRE is in a dual relationship with covariate balancing. TMLE is a method for constructing regression function estimators such that the leading bias term becomes zero. Nearest Neighbor Matching is equivalent to Least Squares Density Ratio Estimation and Riesz Regression.

☆ Clone Deterministic 3D Worlds with Geometrically-Regularized World Models

Zaishuo Xia, Yukuan Lu, Xinyi Li, Yifan Xu, Yubei Chen

A world model is an internal model that simulates how the world evolves. Given past observations and actions, it predicts the future of both the embodied agent and its environment. Accurate world models are essential for enabling agents to think, plan, and reason effectively in complex, dynamic settings. Despite rapid progress, current world models remain brittle and degrade over long horizons. We argue that a central cause is representation quality: exteroceptive inputs (e.g., images) are high-dimensional, and lossy or entangled latents make dynamics learning unnecessarily hard. We therefore ask whether improving representation learning alone can substantially improve world-model performance. In this work, we take a step toward building a truly accurate world model by addressing a fundamental yet open problem: constructing a model that can fully clone and overfit to a deterministic 3D world. We propose Geometrically-Regularized World Models (GRWM), which enforces that consecutive points along a natural sensory trajectory remain close in latent representation space. This approach yields significantly improved latent representations that align closely with the true topology of the environment. GRWM is plug-and-play, requires only minimal architectural modification, scales with trajectory length, and is compatible with diverse latent generative backbones. Across deterministic 3D settings and long-horizon prediction tasks, GRWM significantly increases rollout fidelity and stability. Analyses show that its benefits stem from learning a latent manifold with superior geometric structure. These findings support a clear takeaway: improving representation learning is a direct and useful path to robust world models, delivering reliable long-horizon predictions without enlarging the dynamics module.

☆ Surpassing state of the art on AMD area estimation from RGB fundus images through careful selection of U-Net architectures and loss functions for class imbalance

Valentyna Starodub, Mantas Lukoševičius

Age-related macular degeneration (AMD) is one of the leading causes of irreversible vision impairment in people over the age of 60. This research focuses on semantic segmentation for AMD lesion detection in RGB fundus images, a non-invasive and cost-effective imaging technique. The results of the ADAM challenge - the most comprehensive AMD detection from RGB fundus images research competition and open dataset to date - serve as a benchmark for our evaluation. Taking the U-Net connectivity as a base of our framework, we evaluate and compare several approaches to improve the segmentation model's architecture and training pipeline, including pre-processing techniques, encoder (backbone) deep network types of varying complexity, and specialized loss functions to mitigate class imbalances on image and pixel levels. The main outcome of this research is the final configuration of the AMD detection framework, which outperforms all the prior ADAM challenge submissions on the multi-class segmentation of different AMD lesion types in non-invasive RGB fundus images. The source code used to conduct the experiments presented in this paper is made freely available.

☆ Pre-trained Forecasting Models: Strong Zero-Shot Feature Extractors for Time Series Classification NeurIPS 2025

Andreas Auer, Daniel Klotz, Sebastinan Böck, Sepp Hochreiter

Recent research on time series foundation models has primarily focused on forecasting, leaving it unclear how generalizable their learned representations are. In this study, we examine whether frozen pre-trained forecasting models can provide effective representations for classification. To this end, we compare different representation extraction strategies and introduce two model-agnostic embedding augmentations. Our experiments show that the best forecasting models achieve classification accuracy that matches or even surpasses that of state-of-the-art models pre-trained specifically for classification. Moreover, we observe a positive correlation between forecasting and classification performance. These findings challenge the assumption that task-specific pre-training is necessary, and suggest that learning to forecast may provide a powerful route toward constructing general-purpose time series foundation models.

comment: NeurIPS 2025 Workshop on Recent Advances in Time Series Foundation Models (BERT2S)

☆ Faithful and Fast Influence Function via Advanced Sampling

Jungyeon Koh, Hyeonsu Lyu, Jonggyu Jang, Hyun Jong Yang

How can we explain the influence of training data on black-box models? Influence functions (IFs) offer a post-hoc solution by utilizing gradients and Hessians. However, computing the Hessian for an entire dataset is resource-intensive, necessitating a feasible alternative. A common approach involves randomly sampling a small subset of the training data, but this method often results in highly inconsistent IF estimates due to the high variance in sample configurations. To address this, we propose two advanced sampling techniques based on features and logits. These samplers select a small yet representative subset of the entire dataset by considering the stochastic distribution of features or logits, thereby enhancing the accuracy of IF estimations. We validate our approach through class removal experiments, a typical application of IFs, using the F1-score to measure how effectively the model forgets the removed class while maintaining inference consistency on the remaining classes. Our method reduces computation time by 30.1% and memory usage by 42.2%, or improves the F1-score by 2.5% compared to the baseline.

☆ STaMP: Sequence Transformation and Mixed Precision for Low-Precision Activation Quantization

Marco Federici, Riccardo Del Chiaro, Boris van Breugel, Paul Whatmough, Markus Nagel

Quantization is the key method for reducing inference latency, power and memory footprint of generative AI models. However, accuracy often degrades sharply when activations are quantized below eight bits. Recent work suggests that invertible linear transformations (e.g. rotations) can aid quantization, by reparameterizing feature channels and weights. In this paper, we propose \textit{Sequence Transformation and Mixed Precision} (STaMP) quantization, a novel strategy that applies linear transformations along the \textit{sequence} dimension to exploit the strong local correlation in language and visual data. By keeping a small number of tokens in each intermediate activation at higher precision, we can maintain model accuracy at lower (average) activations bit-widths. We evaluate STaMP on recent LVM and LLM architectures, demonstrating that it significantly improves low bit width activation quantization and complements established activation and weight quantization methods including recent feature transformations.

comment: 10 pages main text, 8 pages supplementary material

☆ SteerVLM: Robust Model Control through Lightweight Activation Steering for Vision Language Models

Anushka Sivakumar, Andrew Zhang, Zaber Hakim, Chris Thomas

This work introduces SteerVLM, a lightweight steering module designed to guide Vision-Language Models (VLMs) towards outputs that better adhere to desired instructions. Our approach learns from the latent embeddings of paired prompts encoding target and converse behaviors to dynamically adjust activations connecting the language modality with image context. This allows for fine-grained, inference-time control over complex output semantics without modifying model weights while preserving performance on off-target tasks. Our steering module requires learning parameters equal to 0.14% of the original VLM's size. Our steering module gains model control through dimension-wise activation modulation and adaptive steering across layers without requiring pre-extracted static vectors or manual tuning of intervention points. Furthermore, we introduce VNIA (Visual Narrative Intent Alignment), a multimodal dataset specifically created to facilitate the development and evaluation of VLM steering techniques. Our method outperforms existing intervention techniques on steering and hallucination mitigation benchmarks for VLMs and proposes a robust solution for multimodal model control through activation engineering.

☆ The Oversight Game: Learning to Cooperatively Balance an AI Agent's Safety and Autonomy

William Overman, Mohsen Bayati

As increasingly capable agents are deployed, a central safety question is how to retain meaningful human control without modifying the underlying system. We study a minimal control interface where an agent chooses whether to act autonomously (play) or defer (ask), while a human simultaneously chooses whether to be permissive (trust) or to engage in oversight (oversee). If the agent defers, the human's choice determines the outcome, potentially leading to a corrective action or a system shutdown. We model this interaction as a two-player Markov Game. Our analysis focuses on cases where this game qualifies as a Markov Potential Game (MPG), a class of games where we can provide an alignment guarantee: under a structural assumption on the human's value function, any decision by the agent to act more autonomously that benefits itself cannot harm the human's value. We also analyze extensions to this MPG framework. Theoretically, this perspective provides conditions for a specific form of intrinsic alignment. If the reward structures of the human-agent game meet these conditions, we have a formal guarantee that the agent improving its own outcome will not harm the human's. Practically, this model motivates a transparent control layer with predictable incentives where the agent learns to defer when risky and act when safe, while its pretrained policy and the environment's reward structure remain untouched. Our gridworld simulation shows that through independent learning, the agent and human discover their optimal oversight roles. The agent learns to ask when uncertain and the human learns when to oversee, leading to an emergent collaboration that avoids safety violations introduced post-training. This demonstrates a practical method for making misaligned models safer after deployment.

☆ Deep sequence models tend to memorize geometrically; it is unclear why

Shahriar Noroozizadeh, Vaishnavh Nagarajan, Elan Rosenfeld, Sanjiv Kumar

In sequence modeling, the parametric memory of atomic facts has been predominantly abstracted as a brute-force lookup of co-occurrences between entities. We contrast this associative view against a geometric view of how memory is stored. We begin by isolating a clean and analyzable instance of Transformer reasoning that is incompatible with memory as strictly a storage of the local co-occurrences specified during training. Instead, the model must have somehow synthesized its own geometry of atomic facts, encoding global relationships between all entities, including non-co-occurring ones. This in turn has simplified a hard reasoning task involving an $\ell$-fold composition into an easy-to-learn 1-step geometric task. From this phenomenon, we extract fundamental aspects of neural embedding geometries that are hard to explain. We argue that the rise of such a geometry, despite optimizing over mere local associations, cannot be straightforwardly attributed to typical architectural or optimizational pressures. Counterintuitively, an elegant geometry is learned even when it is not more succinct than a brute-force lookup of associations. Then, by analyzing a connection to Node2Vec, we demonstrate how the geometry stems from a spectral bias that -- in contrast to prevailing theories -- indeed arises naturally despite the lack of various pressures. This analysis also points to practitioners a visible headroom to make Transformer memory more strongly geometric. We hope the geometric view of parametric memory encourages revisiting the default intuitions that guide researchers in areas like knowledge acquisition, capacity, discovery and unlearning.

☆ Bridging the Gap between Empirical Welfare Maximization and Conditional Average Treatment Effect Estimation in Policy Learning

Masahiro Kato

The goal of policy learning is to train a policy function that recommends a treatment given covariates to maximize population welfare. There are two major approaches in policy learning: the empirical welfare maximization (EWM) approach and the plug-in approach. The EWM approach is analogous to a classification problem, where one first builds an estimator of the population welfare, which is a functional of policy functions, and then trains a policy by maximizing the estimated welfare. In contrast, the plug-in approach is based on regression, where one first estimates the conditional average treatment effect (CATE) and then recommends the treatment with the highest estimated outcome. This study bridges the gap between the two approaches by showing that both are based on essentially the same optimization problem. In particular, we prove an exact equivalence between EWM and least squares over a reparameterization of the policy class. As a consequence, the two approaches are interchangeable in several respects and share the same theoretical guarantees under common conditions. Leveraging this equivalence, we propose a novel regularization method for policy learning. Our findings yield a convex and computationally efficient training procedure that avoids the NP-hard combinatorial step typically required in EWM.

☆ Non-Convex Over-the-Air Heterogeneous Federated Learning: A Bias-Variance Trade-off

Muhammad Faraz Ul Abrar, Nicolò Michelusi

Over-the-air (OTA) federated learning (FL) has been well recognized as a scalable paradigm that exploits the waveform superposition of the wireless multiple-access channel to aggregate model updates in a single use. Existing OTA-FL designs largely enforce zero-bias model updates by either assuming \emph{homogeneous} wireless conditions (equal path loss across devices) or forcing zero-bias updates to guarantee convergence. Under \emph{heterogeneous} wireless scenarios, however, such designs are constrained by the weakest device and inflate the update variance. Moreover, prior analyses of biased OTA-FL largely address convex objectives, while most modern AI models are highly non-convex. Motivated by these gaps, we study OTA-FL with stochastic gradient descent (SGD) for general smooth non-convex objectives under wireless heterogeneity. We develop novel OTA-FL SGD updates that allow a structured, time-invariant model bias while facilitating reduced variance updates. We derive a finite-time stationarity bound (expected time average squared gradient norm) that explicitly reveals a bias-variance trade-off. To optimize this trade-off, we pose a non-convex joint OTA power-control design and develop an efficient successive convex approximation (SCA) algorithm that requires only statistical CSI at the base station. Experiments on a non-convex image classification task validate the approach: the SCA-based design accelerates convergence via an optimized bias and improves generalization over prior OTA-FL baselines.

☆ On Purely Private Covariance Estimation

Tommaso d'Orsi, Gleb Novikov

We present a simple perturbation mechanism for the release of $d$-dimensional covariance matrices $\Sigma$ under pure differential privacy. For large datasets with at least $n\geq d^2/\varepsilon$ elements, our mechanism recovers the provably optimal Frobenius norm error guarantees of \cite{nikolov2023private}, while simultaneously achieving best known error for all other $p$-Schatten norms, with $p\in [1,\infty]$. Our error is information-theoretically optimal for all $p\ge 2$, in particular, our mechanism is the first purely private covariance estimator that achieves optimal error in spectral norm. For small datasets $n< d^2/\varepsilon$, we further show that by projecting the output onto the nuclear norm ball of appropriate radius, our algorithm achieves the optimal Frobenius norm error $O(\sqrt{d\;\text{Tr}(\Sigma) /n})$, improving over the known bounds of $O(\sqrt{d/n})$ of \cite{nikolov2023private} and ${O}\big(d^{3/4}\sqrt{\text{Tr}(\Sigma)/n}\big)$ of \cite{dong2022differentially}.

comment: equal contribution

☆ LSM-MS2: A Foundation Model Bridging Spectral Identification and Biological Interpretation

Gabriel Asher, Devesh Shah, Amy A. Caudy, Luke Ferro, Lea Amar, Ana S. H. Costa, Thomas Patton, Niall O'Connor, Jennifer M. Campbell, Jack Geremia

A vast majority of mass spectrometry data remains uncharacterized, leaving much of its biological and chemical information untapped. Recent advances in machine learning have begun to address this gap, particularly for tasks such as spectral identification in tandem mass spectrometry data. Here, we present the latest generation of LSM-MS2, a large-scale deep learning foundation model trained on millions of spectra to learn a semantic chemical space. LSM-MS2 achieves state-of-the-art performance in spectral identification, improving on existing methods by 30% in accuracy of identifying challenging isomeric compounds, yielding 42% more correct identifications in complex biological samples, and maintaining robustness under low-concentration conditions. Furthermore, LSM-MS2 produces rich spectral embeddings that enable direct biological interpretation from minimal downstream data, successfully differentiating disease states and predicting clinical outcomes across diverse translational applications.

☆ On the limitation of evaluating machine unlearning using only a single training seed

Jamie Lanyon, Axel Finke, Petros Andreou, Georgina Cosma

Machine unlearning (MU) aims to remove the influence of certain data points from a trained model without costly retraining. Most practical MU algorithms are only approximate and their performance can only be assessed empirically. Care must therefore be taken to make empirical comparisons as representative as possible. A common practice is to run the MU algorithm multiple times independently starting from the same trained model. In this work, we demonstrate that this practice can give highly non-representative results because -- even for the same architecture and same dataset -- some MU methods can be highly sensitive to the choice of random number seed used for model training. We therefore recommend that empirical comphttps://info.arxiv.org/help/prep#commentsarisons of MU algorithms should also reflect the variability across different model training seeds.

comment: mini paper, 2 figures

☆ An All-Reduce Compatible Top-K Compressor for Communication-Efficient Distributed Learning

Chuyan Chen, Chenyang Ma, Zhangxin Li, Yutong He, Yanjie Dong, Kun Yuan

Communication remains a central bottleneck in large-scale distributed machine learning, and gradient sparsification has emerged as a promising strategy to alleviate this challenge. However, existing gradient compressors face notable limitations: Rand-$K$\ discards structural information and performs poorly in practice, while Top-$K$\ preserves informative entries but loses the contraction property and requires costly All-Gather operations. In this paper, we propose ARC-Top-$K$, an {All-Reduce}-Compatible Top-$K$ compressor that aligns sparsity patterns across nodes using a lightweight sketch of the gradient, enabling index-free All-Reduce while preserving globally significant information. ARC-Top-$K$\ is provably contractive and, when combined with momentum error feedback (EF21M), achieves linear speedup and sharper convergence rates than the original EF21M under standard assumptions. Empirically, ARC-Top-$K$\ matches the accuracy of Top-$K$\ while reducing wall-clock training time by up to 60.7\%, offering an efficient and scalable solution that combines the robustness of Rand-$K$\ with the strong performance of Top-$K$.

comment: 8 pages, 2 figures

☆ Value Drifts: Tracing Value Alignment During LLM Post-Training

Mehar Bhatia, Shravan Nayak, Gaurav Kamath, Marius Mosbach, Karolina Stańczak, Vered Shwartz, Siva Reddy

As LLMs occupy an increasingly important role in society, they are more and more confronted with questions that require them not only to draw on their general knowledge but also to align with certain human value systems. Therefore, studying the alignment of LLMs with human values has become a crucial field of inquiry. Prior work, however, mostly focuses on evaluating the alignment of fully trained models, overlooking the training dynamics by which models learn to express human values. In this work, we investigate how and at which stage value alignment arises during the course of a model's post-training. Our analysis disentangles the effects of post-training algorithms and datasets, measuring both the magnitude and time of value drifts during training. Experimenting with Llama-3 and Qwen-3 models of different sizes and popular supervised fine-tuning (SFT) and preference optimization datasets and algorithms, we find that the SFT phase generally establishes a model's values, and subsequent preference optimization rarely re-aligns these values. Furthermore, using a synthetic preference dataset that enables controlled manipulation of values, we find that different preference optimization algorithms lead to different value alignment outcomes, even when preference data is held constant. Our findings provide actionable insights into how values are learned during post-training and help to inform data curation, as well as the selection of models and algorithms for preference optimization to improve model alignment to human values.

☆ Budgeted Multiple-Expert Deferral

Giulia DeSalvo, Clara Mohri, Mehryar Mohri, Yutao Zhong

Learning to defer uncertain predictions to costly experts offers a powerful strategy for improving the accuracy and efficiency of machine learning systems. However, standard training procedures for deferral algorithms typically require querying all experts for every training instance, an approach that becomes prohibitively expensive when expert queries incur significant computational or resource costs. This undermines the core goal of deferral: to limit unnecessary expert usage. To overcome this challenge, we introduce the budgeted deferral framework, which aims to train effective deferral algorithms while minimizing expert query costs during training. We propose new algorithms for both two-stage and single-stage multiple-expert deferral settings that selectively query only a subset of experts per training example. While inspired by active learning, our setting is fundamentally different: labels are already known, and the core challenge is to decide which experts to query in order to balance cost and predictive performance. We establish theoretical guarantees for both of our algorithms, including generalization bounds and label complexity analyses. Empirical results across several domains show that our algorithms substantially reduce training costs without sacrificing prediction accuracy, demonstrating the practical value of our budget-aware deferral algorithms.

☆ How Regularization Terms Make Invertible Neural Networks Bayesian Point Estimators

Nick Heilenkötter

Can regularization terms in the training of invertible neural networks lead to known Bayesian point estimators in reconstruction? Invertible networks are attractive for inverse problems due to their inherent stability and interpretability. Recently, optimization strategies for invertible neural networks that approximate either a reconstruction map or the forward operator have been studied from a Bayesian perspective, but each has limitations. To address this, we introduce and analyze two regularization terms for the network training that, upon inversion of the network, recover properties of classical Bayesian point estimators: while the first can be connected to the posterior mean, the second resembles the MAP estimator. Our theoretical analysis characterizes how each loss shapes both the learned forward operator and its inverse reconstruction map. Numerical experiments support our findings and demonstrate how these loss-term regularizers introduce data-dependence in a stable and interpretable way.

comment: Preprint, under review

☆ Assessment of the conditional exchangeability assumption in causal machine learning models: a simulation study

Gerard T. Portela, Jason B. Gibbons, Sebastian Schneeweiss, Rishi J. Desai

Observational studies developing causal machine learning (ML) models for the prediction of individualized treatment effects (ITEs) seldom conduct empirical evaluations to assess the conditional exchangeability assumption. We aimed to evaluate the performance of these models under conditional exchangeability violations and the utility of negative control outcomes (NCOs) as a diagnostic. We conducted a simulation study to examine confounding bias in ITE estimates generated by causal forest and X-learner models under varying conditions, including the presence or absence of true heterogeneity. We simulated data to reflect real-world scenarios with differing levels of confounding, sample size, and NCO confounding structures. We then estimated and compared subgroup-level treatment effects on the primary outcome and NCOs across settings with and without unmeasured confounding. When conditional exchangeability was violated, causal forest and X-learner models failed to recover true treatment effect heterogeneity and, in some cases, falsely indicated heterogeneity when there was none. NCOs successfully identified subgroups affected by unmeasured confounding. Even when NCOs did not perfectly satisfy its ideal assumptions, it remained informative, flagging potential bias in subgroup level estimates, though not always pinpointing the subgroup with the largest confounding. Violations of conditional exchangeability substantially limit the validity of ITE estimates from causal ML models in routinely collected observational data. NCOs serve a useful empirical diagnostic tool for detecting subgroup-specific unmeasured confounding and should be incorporated into causal ML workflows to support the credibility of individualized inference.

☆ Kimi Linear: An Expressive, Efficient Attention Architecture

Kimi Team, Yu Zhang, Zongyu Lin, Xingcheng Yao, Jiaxi Hu, Fanqing Meng, Chengyin Liu, Xin Men, Songlin Yang, Zhiyuan Li, Wentao Li, Enzhe Lu, Weizhou Liu, Yanru Chen, Weixin Xu, Longhui Yu, Yejie Wang, Yu Fan, Longguang Zhong, Enming Yuan, Dehao Zhang, Yizhi Zhang, T. Y. Liu, Haiming Wang, Shengjun Fang, Weiran He, Shaowei Liu, Yiwei Li, Jianlin Su, Jiezhong Qiu, Bo Pang, Junjie Yan, Zhejun Jiang, Weixiao Huang, Bohong Yin, Jiacheng You, Chu Wei, Zhengtao Wang, Chao Hong, Yutian Chen, Guanduo Chen, Yucheng Wang, Huabin Zheng, Feng Wang, Yibo Liu, Mengnan Dong, Zheng Zhang, Siyuan Pan, Wenhao Wu, Yuhao Wu, Longyu Guan, Jiawen Tao, Guohong Fu, Xinran Xu, Yuzhi Wang, Guokun Lai, Yuxin Wu, Xinyu Zhou, Zhilin Yang, Yulun Du

We introduce Kimi Linear, a hybrid linear attention architecture that, for the first time, outperforms full attention under fair comparisons across various scenarios -- including short-context, long-context, and reinforcement learning (RL) scaling regimes. At its core lies Kimi Delta Attention (KDA), an expressive linear attention module that extends Gated DeltaNet with a finer-grained gating mechanism, enabling more effective use of limited finite-state RNN memory. Our bespoke chunkwise algorithm achieves high hardware efficiency through a specialized variant of the Diagonal-Plus-Low-Rank (DPLR) transition matrices, which substantially reduces computation compared to the general DPLR formulation while remaining more consistent with the classical delta rule. We pretrain a Kimi Linear model with 3B activated parameters and 48B total parameters, based on a layerwise hybrid of KDA and Multi-Head Latent Attention (MLA). Our experiments show that with an identical training recipe, Kimi Linear outperforms full MLA with a sizeable margin across all evaluated tasks, while reducing KV cache usage by up to 75% and achieving up to 6 times decoding throughput for a 1M context. These results demonstrate that Kimi Linear can be a drop-in replacement for full attention architectures with superior performance and efficiency, including tasks with longer input and output lengths. To support further research, we open-source the KDA kernel and vLLM implementations, and release the pre-trained and instruction-tuned model checkpoints.

comment: Kimi Linear tech report

☆ LoRAQuant: Mixed-Precision Quantization of LoRA to Ultra-Low Bits

Amir Reza Mirzaei, Yuqiao Wen, Yanshuai Cao, Lili Mou

Low-Rank Adaptation (LoRA) has become a popular technique for parameter-efficient fine-tuning of large language models (LLMs). In many real-world scenarios, multiple adapters are loaded simultaneously to enable LLM customization for personalized user experiences or to support a diverse range of tasks. Although each adapter is lightweight in isolation, their aggregate cost becomes substantial at scale. To address this, we propose LoRAQuant, a mixed-precision post-training quantization method tailored to LoRA. Specifically, LoRAQuant reparameterizes each adapter by singular value decomposition (SVD) to concentrate the most important information into specific rows and columns. This makes it possible to quantize the important components to higher precision, while quantizing the rest to ultra-low bitwidth. We conduct comprehensive experiments with LLaMA 2-7B, LLaMA 2-13B, and Mistral 7B models on mathematical reasoning, coding, and summarization tasks. Results show that our LoRAQuant uses significantly lower bits than other quantization methods, but achieves comparable or even higher performance.

☆ FlowQ-Net: A Generative Framework for Automated Quantum Circuit Design

Jun Dai, Michael Rizvi-Martel, Guillaume Rabusseau

Designing efficient quantum circuits is a central bottleneck to exploring the potential of quantum computing, particularly for noisy intermediate-scale quantum (NISQ) devices, where circuit efficiency and resilience to errors are paramount. The search space of gate sequences grows combinatorially, and handcrafted templates often waste scarce qubit and depth budgets. We introduce \textsc{FlowQ-Net} (Flow-based Quantum design Network), a generative framework for automated quantum circuit synthesis based on Generative Flow Networks (GFlowNets). This framework learns a stochastic policy to construct circuits sequentially, sampling them in proportion to a flexible, user-defined reward function that can encode multiple design objectives such as performance, depth, and gate count. This approach uniquely enables the generation of a diverse ensemble of high-quality circuits, moving beyond single-solution optimization. We demonstrate the efficacy of \textsc{FlowQ-Net} through an extensive set of simulations. We apply our method to Variational Quantum Algorithm (VQA) ansatz design for molecular ground state estimation, Max-Cut, and image classification, key challenges in near-term quantum computing. Circuits designed by \textsc{FlowQ-Net} achieve significant improvements, yielding circuits that are 10$\times$-30$\times$ more compact in terms of parameters, gates, and depth compared to commonly used unitary baselines, without compromising accuracy. This trend holds even when subjected to error profiles from real-world quantum devices. Our results underline the potential of generative models as a general-purpose methodology for automated quantum circuit design, offering a promising path towards more efficient quantum algorithms and accelerating scientific discovery in the quantum domain.

☆ Tight Differentially Private PCA via Matrix Coherence

Tommaso d'Orsi, Gleb Novikov

We revisit the task of computing the span of the top $r$ singular vectors $u_1, \ldots, u_r$ of a matrix under differential privacy. We show that a simple and efficient algorithm -- based on singular value decomposition and standard perturbation mechanisms -- returns a private rank-$r$ approximation whose error depends only on the \emph{rank-$r$ coherence} of $u_1, \ldots, u_r$ and the spectral gap $\sigma_r - \sigma_{r+1}$. This resolves a question posed by Hardt and Roth~\cite{hardt2013beyond}. Our estimator outperforms the state of the art -- significantly so in some regimes. In particular, we show that in the dense setting, it achieves the same guarantees for single-spike PCA in the Wishart model as those attained by optimal non-private algorithms, whereas prior private algorithms failed to do so. In addition, we prove that (rank-$r$) coherence does not increase under Gaussian perturbations. This implies that any estimator based on the Gaussian mechanism -- including ours -- preserves the coherence of the input. We conjecture that similar behavior holds for other structured models, including planted problems in graphs. We also explore applications of coherence to graph problems. In particular, we present a differentially private algorithm for Max-Cut and other constraint satisfaction problems under low coherence assumptions.

comment: SODA 2026; equal contribution

☆ Action-Driven Processes for Continuous-Time Control

Ruimin He, Shaowei Lin

At the heart of reinforcement learning are actions -- decisions made in response to observations of the environment. Actions are equally fundamental in the modeling of stochastic processes, as they trigger discontinuous state transitions and enable the flow of information through large, complex systems. In this paper, we unify the perspectives of stochastic processes and reinforcement learning through action-driven processes, and illustrate their application to spiking neural networks. Leveraging ideas from control-as-inference, we show that minimizing the Kullback-Leibler divergence between a policy-driven true distribution and a reward-driven model distribution for a suitably defined action-driven process is equivalent to maximum entropy reinforcement learning.

☆ Heuristic Adaptation of Potentially Misspecified Domain Support for Likelihood-Free Inference in Stochastic Dynamical Systems

Georgios Kamaras, Craig Innes, Subramanian Ramamoorthy

In robotics, likelihood-free inference (LFI) can provide the domain distribution that adapts a learnt agent in a parametric set of deployment conditions. LFI assumes an arbitrary support for sampling, which remains constant as the initial generic prior is iteratively refined to more descriptive posteriors. However, a potentially misspecified support can lead to suboptimal, yet falsely certain, posteriors. To address this issue, we propose three heuristic LFI variants: EDGE, MODE, and CENTRE. Each interprets the posterior mode shift over inference steps in its own way and, when integrated into an LFI step, adapts the support alongside posterior inference. We first expose the support misspecification issue and evaluate our heuristics using stochastic dynamical benchmarks. We then evaluate the impact of heuristic support adaptation on parameter inference and policy learning for a dynamic deformable linear object (DLO) manipulation task. Inference results in a finer length and stiffness classification for a parametric set of DLOs. When the resulting posteriors are used as domain distributions for sim-based policy learning, they lead to more robust object-centric agent performance.

☆ Hybrid DQN-TD3 Reinforcement Learning for Autonomous Navigation in Dynamic Environments

Xiaoyi He, Danggui Chen, Zhenshuo Zhang, Zimeng Bai

This paper presents a hierarchical path-planning and control framework that combines a high-level Deep Q-Network (DQN) for discrete sub-goal selection with a low-level Twin Delayed Deep Deterministic Policy Gradient (TD3) controller for continuous actuation. The high-level module selects behaviors and sub-goals; the low-level module executes smooth velocity commands. We design a practical reward shaping scheme (direction, distance, obstacle avoidance, action smoothness, collision penalty, time penalty, and progress), together with a LiDAR-based safety gate that prevents unsafe motions. The system is implemented in ROS + Gazebo (TurtleBot3) and evaluated with PathBench metrics, including success rate, collision rate, path efficiency, and re-planning efficiency, in dynamic and partially observable environments. Experiments show improved success rate and sample efficiency over single-algorithm baselines (DQN or TD3 alone) and rule-based planners, with better generalization to unseen obstacle configurations and reduced abrupt control changes. Code and evaluation scripts are available at the project repository.

comment: 6 pages, 5 figures; ROS+Gazebo (TurtleBot3) implementation; evaluation with PathBench metrics; code (primary): https://github.com/MayaCHEN-github/HierarchicalRL-robot-navigation; mirror (for reproducibility): https://github.com/ShowyHe/DRL-robot-navigation

☆ Curly Flow Matching for Learning Non-gradient Field Dynamics NeurIPS 2025

Katarina Petrović, Lazar Atanackovic, Viggo Moro, Kacper Kapuśniak, İsmail İlkan Ceylan, Michael Bronstein, Avishek Joey Bose, Alexander Tong

Modeling the transport dynamics of natural processes from population-level observations is a ubiquitous problem in the natural sciences. Such models rely on key assumptions about the underlying process in order to enable faithful learning of governing dynamics that mimic the actual system behavior. The de facto assumption in current approaches relies on the principle of least action that results in gradient field dynamics and leads to trajectories minimizing an energy functional between two probability measures. However, many real-world systems, such as cell cycles in single-cell RNA, are known to exhibit non-gradient, periodic behavior, which fundamentally cannot be captured by current state-of-the-art methods such as flow and bridge matching. In this paper, we introduce Curly Flow Matching (Curly-FM), a novel approach that is capable of learning non-gradient field dynamics by designing and solving a Schr\"odinger bridge problem with a non-zero drift reference process -- in stark contrast to typical zero-drift reference processes -- which is constructed using inferred velocities in addition to population snapshot data. We showcase Curly-FM by solving the trajectory inference problems for single cells, computational fluid dynamics, and ocean currents with approximate velocities. We demonstrate that Curly-FM can learn trajectories that better match both the reference process and population marginals. Curly-FM expands flow matching models beyond the modeling of populations and towards the modeling of known periodic behavior in physical systems. Our code repository is accessible at: https://github.com/kpetrovicc/curly-flow-matching.git

comment: Accepted to NeurIPS 2025

☆ MSAD: A Deep Dive into Model Selection for Time series Anomaly Detection VLDB

Emmanouil Sylligardos, John Paparrizos, Themis Palpanas, Pierre Senellart, Paul Boniol

Anomaly detection is a fundamental task for time series analytics with important implications for the downstream performance of many applications. Despite increasing academic interest and the large number of methods proposed in the literature, recent benchmarks and evaluation studies demonstrated that no overall best anomaly detection methods exist when applied to very heterogeneous time series datasets. Therefore, the only scalable and viable solution to solve anomaly detection over very different time series collected from diverse domains is to propose a model selection method that will select, based on time series characteristics, the best anomaly detection methods to run. Existing AutoML solutions are, unfortunately, not directly applicable to time series anomaly detection, and no evaluation of time series-based approaches for model selection exists. Towards that direction, this paper studies the performance of time series classification methods used as model selection for anomaly detection. In total, we evaluate 234 model configurations derived from 16 base classifiers across more than 1980 time series, and we propose the first extensive experimental evaluation of time series classification as model selection for anomaly detection. Our results demonstrate that model selection methods outperform every single anomaly detection method while being in the same order of magnitude regarding execution time. This evaluation is the first step to demonstrate the accuracy and efficiency of time series classification algorithms for anomaly detection, and represents a strong baseline that can then be used to guide the model selection step in general AutoML pipelines. Preprint version of an article accepted at the VLDB Journal.

comment: 25 pages, 13 figures, VLDB Journal

☆ Omnipresent Yet Overlooked: Heat Kernels in Combinatorial Bayesian Optimization

Colin Doumont, Victor Picheny, Viacheslav Borovitskiy, Henry Moss

Bayesian Optimization (BO) has the potential to solve various combinatorial tasks, ranging from materials science to neural architecture search. However, BO requires specialized kernels to effectively model combinatorial domains. Recent efforts have introduced several combinatorial kernels, but the relationships among them are not well understood. To bridge this gap, we develop a unifying framework based on heat kernels, which we derive in a systematic way and express as simple closed-form expressions. Using this framework, we prove that many successful combinatorial kernels are either related or equivalent to heat kernels, and validate this theoretical claim in our experiments. Moreover, our analysis confirms and extends the results presented in Bounce: certain algorithms' performance decreases substantially when the unknown optima of the function do not have a certain structure. In contrast, heat kernels are not sensitive to the location of the optima. Lastly, we show that a fast and simple pipeline, relying on heat kernels, is able to achieve state-of-the-art results, matching or even outperforming certain slow or complex algorithms.

☆ Aeolus: A Multi-structural Flight Delay Dataset

Lin Xu, Xinyun Yuan, Yuxuan Liang, Suwan Yin, Yuankai Wu

We introduce Aeolus, a large-scale Multi-modal Flight Delay Dataset designed to advance research on flight delay prediction and support the development of foundation models for tabular data. Existing datasets in this domain are typically limited to flat tabular structures and fail to capture the spatiotemporal dynamics inherent in delay propagation. Aeolus addresses this limitation by providing three aligned modalities: (i) a tabular dataset with rich operational, meteorological, and airportlevel features for over 50 million flights; (ii) a flight chain module that models delay propagation along sequential flight legs, capturing upstream and downstream dependencies; and (iii) a flight network graph that encodes shared aircraft, crew, and airport resource connections, enabling cross-flight relational reasoning. The dataset is carefully constructed with temporal splits, comprehensive features, and strict leakage prevention to support realistic and reproducible machine learning evaluation. Aeolus supports a broad range of tasks, including regression, classification, temporal structure modeling, and graph learning, serving as a unified benchmark across tabular, sequential, and graph modalities. We release baseline experiments and preprocessing tools to facilitate adoption. Aeolus fills a key gap for both domain-specific modeling and general-purpose structured data research.Our source code and data can be accessed at https://github.com/Flnny/Delay-data

☆ CYPRESS: Crop Yield Prediction via Regression on Prithvi's Encoder for Satellite Sensing

Shayan Nejadshamsi, Yuanyuan Zhang, Shadi Zaki, Brock Porth, Lysa Porth, Vahab Khoshdel

Accurate and timely crop yield prediction is crucial for global food security and modern agricultural management. Traditional methods often lack the scalability and granularity required for precision farming. This paper introduces CYPRESS (Crop Yield Prediction via Regression on Prithvi's Encoder for Satellite Sensing), a deep learning model designed for high-resolution, intra-field canola yield prediction. CYPRESS leverages a pre-trained, large-scale geospatial foundation model (Prithvi-EO-2.0-600M) and adapts it for a continuous regression task, transforming multi-temporal satellite imagery into dense, pixel-level yield maps. Evaluated on a comprehensive dataset from the Canadian Prairies, CYPRESS demonstrates superior performance over existing deep learning-based yield prediction models, highlighting the effectiveness of fine-tuning foundation models for specialized agricultural applications. By providing a continuous, high-resolution output, CYPRESS offers a more actionable tool for precision agriculture than conventional classification or county-level aggregation methods. This work validates a novel approach that bridges the gap between large-scale Earth observation and on-farm decision-making, offering a scalable solution for detailed agricultural monitoring.

☆ Wasserstein Regression as a Variational Approximation of Probabilistic Trajectories through the Bernstein Basis

Maksim Maslov, Alexander Kugaevskikh, Matthew Ivanov

This paper considers the problem of regression over distributions, which is becoming increasingly important in machine learning. Existing approaches often ignore the geometry of the probability space or are computationally expensive. To overcome these limitations, a new method is proposed that combines the parameterization of probability trajectories using a Bernstein basis and the minimization of the Wasserstein distance between distributions. The key idea is to model a conditional distribution as a smooth probability trajectory defined by a weighted sum of Gaussian components whose parameters -- the mean and covariance -- are functions of the input variable constructed using Bernstein polynomials. The loss function is the averaged squared Wasserstein distance between the predicted Gaussian distributions and the empirical data, which takes into account the geometry of the distributions. An autodiff-based optimization method is used to train the model. Experiments on synthetic datasets that include complex trajectories demonstrated that the proposed method provides competitive approximation quality in terms of the Wasserstein distance, Energy Distance, and RMSE metrics, especially in cases of pronounced nonlinearity. The model demonstrates trajectory smoothness that is better than or comparable to alternatives and robustness to changes in data structure, while maintaining high interpretability due to explicit parameterization via control points. The developed approach represents a balanced solution that combines geometric accuracy, computational practicality, and interpretability. Prospects for further research include extending the method to non-Gaussian distributions, applying entropy regularization to speed up computations, and adapting the approach to working with high-dimensional data for approximating surfaces and more complex structures.

☆ Hybrid Physical-Neural Simulator for Fast Cosmological Hydrodynamics NeurIPS 2025

Arne Thomsen, Tilman Tröster, François Lanusse

Cosmological field-level inference requires differentiable forward models that solve the challenging dynamics of gas and dark matter under hydrodynamics and gravity. We propose a hybrid approach where gravitational forces are computed using a differentiable particle-mesh solver, while the hydrodynamics are parametrized by a neural network that maps local quantities to an effective pressure field. We demonstrate that our method improves upon alternative approaches, such as an Enthalpy Gradient Descent baseline, both at the field and summary-statistic level. The approach is furthermore highly data efficient, with a single reference simulation of cosmological structure formation being sufficient to constrain the neural pressure model. This opens the door for future applications where the model is fit directly to observational data, rather than a training set of simulations.

comment: Accepted to the NeurIPS 2025 Workshop on Machine Learning and the Physical Sciences

☆ Physics-Informed Mixture Models and Surrogate Models for Precision Additive Manufacturing

Sebastian Basterrech, Shuo Shan, Debabrata Adhikari, Sankhya Mohanty

In this study, we leverage a mixture model learning approach to identify defects in laser-based Additive Manufacturing (AM) processes. By incorporating physics based principles, we also ensure that the model is sensitive to meaningful physical parameter variations. The empirical evaluation was conducted by analyzing real-world data from two AM processes: Directed Energy Deposition and Laser Powder Bed Fusion. In addition, we also studied the performance of the developed framework over public datasets with different alloy type and experimental parameter information. The results show the potential of physics-guided mixture models to examine the underlying physical behavior of an AM system.

comment: Five pages, four figures, to be presented at the AI in Science Summit, Denmark, November, 2025

☆ Inference-Cost-Aware Dynamic Tree Construction for Efficient Inference in Large Language Models

Yinrong Hong, Zhiquan Tan, Kai Hu

Large Language Models (LLMs) face significant inference latency challenges stemming from their autoregressive design and large size. To address this, speculative decoding emerges as a solution, enabling the simultaneous generation and validation of multiple tokens. While recent approaches like EAGLE-2 and EAGLE-3 improve speculative decoding using dynamic tree structures, they often neglect the impact of crucial system variables such as GPU devices and batch sizes. Therefore, we introduce a new dynamic tree decoding approach called CAST that takes into account inference costs, including factors such as GPU configurations and batch sizes, to dynamically refine the tree structure. Through comprehensive experimentation across six diverse tasks and utilizing six distinct LLMs, our methodology demonstrates remarkable results, achieving speeds up to 5.2 times faster than conventional decoding methods. Moreover, it generally outperforms existing state-of-the-art techniques from 5% to 20%.

☆ Multiclass Local Calibration With the Jensen-Shannon Distance

Cesare Barbera, Lorenzo Perini, Giovanni De Toni, Andrea Passerini, Andrea Pugnana

Developing trustworthy Machine Learning (ML) models requires their predicted probabilities to be well-calibrated, meaning they should reflect true-class frequencies. Among calibration notions in multiclass classification, strong calibration is the most stringent, as it requires all predicted probabilities to be simultaneously calibrated across all classes. However, existing approaches to multiclass calibration lack a notion of distance among inputs, which makes them vulnerable to proximity bias: predictions in sparse regions of the feature space are systematically miscalibrated. This is especially relevant in high-stakes settings, such as healthcare, where the sparse instances are exactly those most at risk of biased treatment. In this work, we address this main shortcoming by introducing a local perspective on multiclass calibration. First, we formally define multiclass local calibration and establish its relationship with strong calibration. Second, we theoretically analyze the pitfalls of existing evaluation metrics when applied to multiclass local calibration. Third, we propose a practical method for enhancing local calibration in Neural Networks, which enforces alignment between predicted probabilities and local estimates of class frequencies using the Jensen-Shannon distance. Finally, we empirically validate our approach against existing multiclass calibration techniques.

☆ On Measuring Localization of Shortcuts in Deep Networks

Nikita Tsoy, Nikola Konstantinov

Shortcuts, spurious rules that perform well during training but fail to generalize, present a major challenge to the reliability of deep networks (Geirhos et al., 2020). However, the impact of shortcuts on feature representations remains understudied, obstructing the design of principled shortcut-mitigation methods. To overcome this limitation, we investigate the layer-wise localization of shortcuts in deep models. Our novel experiment design quantifies the layer-wise contribution to accuracy degradation caused by a shortcut-inducing skew by counterfactual training on clean and skewed datasets. We employ our design to study shortcuts on CIFAR-10, Waterbirds, and CelebA datasets across VGG, ResNet, DeiT, and ConvNeXt architectures. We find that shortcut learning is not localized in specific layers but distributed throughout the network. Different network parts play different roles in this process: shallow layers predominantly encode spurious features, while deeper layers predominantly forget core features that are predictive on clean data. We also analyze the differences in localization and describe its principal axes of variation. Finally, our analysis of layer-wise shortcut-mitigation strategies suggests the hardness of designing general methods, supporting dataset- and architecture-specific approaches instead.

☆ Boosted Trees on a Diet: Compact Models for Resource-Constrained Devices

Jan Stenkamp, Nina Herrmann, Benjamin Karic, Stefan Oehmcke, Fabian Gieseke

Deploying machine learning models on compute-constrained devices has become a key building block of modern IoT applications. In this work, we present a compression scheme for boosted decision trees, addressing the growing need for lightweight machine learning models. Specifically, we provide techniques for training compact boosted decision tree ensembles that exhibit a reduced memory footprint by rewarding, among other things, the reuse of features and thresholds during training. Our experimental evaluation shows that models achieved the same performance with a compression ratio of 4-16x compared to LightGBM models using an adapted training process and an alternative memory layout. Once deployed, the corresponding IoT devices can operate independently of constant communication or external energy supply, and, thus, autonomously, requiring only minimal computing power and energy. This capability opens the door to a wide range of IoT applications, including remote monitoring, edge analytics, and real-time decision making in isolated or power-limited environments.

☆ Adaptive Inverse Kinematics Framework for Learning Variable-Length Tool Manipulation in Robotics

Prathamesh Kothavale, Sravani Boddepalli

Conventional robots possess a limited understanding of their kinematics and are confined to preprogrammed tasks, hindering their ability to leverage tools efficiently. Driven by the essential components of tool usage - grasping the desired outcome, selecting the most suitable tool, determining optimal tool orientation, and executing precise manipulations - we introduce a pioneering framework. Our novel approach expands the capabilities of the robot's inverse kinematics solver, empowering it to acquire a sequential repertoire of actions using tools of varying lengths. By integrating a simulation-learned action trajectory with the tool, we showcase the practicality of transferring acquired skills from simulation to real-world scenarios through comprehensive experimentation. Remarkably, our extended inverse kinematics solver demonstrates an impressive error rate of less than 1 cm. Furthermore, our trained policy achieves a mean error of 8 cm in simulation. Noteworthy, our model achieves virtually indistinguishable performance when employing two distinct tools of different lengths. This research provides an indication of potential advances in the exploration of all four fundamental aspects of tool usage, enabling robots to master the intricate art of tool manipulation across diverse tasks.

comment: 10 pages, 5 figures. Demonstrates a reinforcement learning framework for adaptive tool manipulation with variable-length extensions

☆ The Structure of Relation Decoding Linear Operators in Large Language Models NeurIPS 2025

Miranda Anna Christ, Adrián Csiszárik, Gergely Becsó, Dániel Varga

This paper investigates the structure of linear operators introduced in Hernandez et al. [2023] that decode specific relational facts in transformer language models. We extend their single-relation findings to a collection of relations and systematically chart their organization. We show that such collections of relation decoders can be highly compressed by simple order-3 tensor networks without significant loss in decoding accuracy. To explain this surprising redundancy, we develop a cross-evaluation protocol, in which we apply each linear decoder operator to the subjects of every other relation. Our results reveal that these linear maps do not encode distinct relations, but extract recurring, coarse-grained semantic properties (e.g., country of capital city and country of food are both in the country-of-X property). This property-centric structure clarifies both the operators' compressibility and highlights why they generalize only to new relations that are semantically close. Our findings thus interpret linear relational decoding in transformer language models as primarily property-based, rather than relation-specific.

comment: NeurIPS 2025 (Spotlight)

☆ A Three-Stage Bayesian Transfer Learning Framework to Improve Predictions in Data-Scarce Domains

Aidan Furlong, Robert Salko, Xingang Zhao, Xu Wu

The use of ML in engineering has grown steadily to support a wide array of applications. Among these methods, deep neural networks have been widely adopted due to their performance and accessibility, but they require large, high-quality datasets. Experimental data are often sparse, noisy, or insufficient to build resilient data-driven models. Transfer learning, which leverages relevant data-abundant source domains to assist learning in data-scarce target domains, has shown efficacy. Parameter transfer, where pretrained weights are reused, is common but degrades under large domain shifts. Domain-adversarial neural networks (DANNs) help address this issue by learning domain-invariant representations, thereby improving transfer under greater domain shifts in a semi-supervised setting. However, DANNs can be unstable during training and lack a native means for uncertainty quantification. This study introduces a fully-supervised three-stage framework, the staged Bayesian domain-adversarial neural network (staged B-DANN), that combines parameter transfer and shared latent space adaptation. In Stage 1, a deterministic feature extractor is trained on the source domain. This feature extractor is then adversarially refined using a DANN in Stage 2. In Stage 3, a Bayesian neural network is built on the adapted feature extractor for fine-tuning on the target domain to handle conditional shifts and yield calibrated uncertainty estimates. This staged B-DANN approach was first validated on a synthetic benchmark, where it was shown to significantly outperform standard transfer techniques. It was then applied to the task of predicting critical heat flux in rectangular channels, leveraging data from tube experiments as the source domain. The results of this study show that the staged B-DANN method can improve predictive accuracy and generalization, potentially assisting other domains in nuclear engineering.

comment: Submitted to Engineering Applications of Artificial Intelligence

☆ Higher-Order Regularization Learning on Hypergraphs

Adrien Weihs, Andrea Bertozzi, Matthew Thorpe

Higher-Order Hypergraph Learning (HOHL) was recently introduced as a principled alternative to classical hypergraph regularization, enforcing higher-order smoothness via powers of multiscale Laplacians induced by the hypergraph structure. Prior work established the well- and ill-posedness of HOHL through an asymptotic consistency analysis in geometric settings. We extend this theoretical foundation by proving the consistency of a truncated version of HOHL and deriving explicit convergence rates when HOHL is used as a regularizer in fully supervised learning. We further demonstrate its strong empirical performance in active learning and in datasets lacking an underlying geometric structure, highlighting HOHL's versatility and robustness across diverse learning settings.

☆ Polybasic Speculative Decoding Through a Theoretical Perspective

Ruilin Wang, Huixia Li, Yuexiao Ma, Xiawu Zheng, Fei Chao, Xuefeng Xiao, Rongrong Ji

Inference latency stands as a critical bottleneck in the large-scale deployment of Large Language Models (LLMs). Speculative decoding methods have recently shown promise in accelerating inference without compromising the output distribution. However, existing work typically relies on a dualistic draft-verify framework and lacks rigorous theoretical grounding. In this paper, we introduce a novel \emph{polybasic} speculative decoding framework, underpinned by a comprehensive theoretical analysis. Specifically, we prove a fundamental theorem that characterizes the optimal inference time for multi-model speculative decoding systems, shedding light on how to extend beyond the dualistic approach to a more general polybasic paradigm. Through our theoretical investigation of multi-model token generation, we expose and optimize the interplay between model capabilities, acceptance lengths, and overall computational cost. Our framework supports both standalone implementation and integration with existing speculative techniques, leading to accelerated performance in practice. Experimental results across multiple model families demonstrate that our approach yields speedup ratios ranging from $3.31\times$ to $4.01\times$ for LLaMA2-Chat 7B, up to $3.87 \times$ for LLaMA3-8B, up to $4.43 \times$ for Vicuna-7B and up to $3.85 \times$ for Qwen2-7B -- all while preserving the original output distribution. We release our theoretical proofs and implementation code to facilitate further investigation into polybasic speculative decoding.

☆ Think Outside the Policy: In-Context Steered Policy Optimization

Hsiu-Yuan Huang, Chenming Tang, Weijie Liu, Saiyong Yang, Yunfang Wu

Existing Reinforcement Learning from Verifiable Rewards (RLVR) methods, such as Group Relative Policy Optimization (GRPO), have achieved remarkable progress in improving the reasoning capabilities of Large Reasoning Models (LRMs). However, they exhibit limited exploration due to reliance on on-policy rollouts where confined to the current policy's distribution, resulting in narrow trajectory diversity. Recent approaches attempt to expand policy coverage by incorporating trajectories generated from stronger expert models, yet this reliance increases computational cost and such advaned models are often inaccessible. To address these issues, we propose In-Context Steered Policy Optimization (ICPO), a unified framework that leverages the inherent in-context learning capability of LRMs to provide expert guidance using existing datasets. ICPO introduces Mixed-Policy GRPO with Implicit Expert Forcing, which expands exploration beyond the current policy distribution without requiring advanced LRM trajectories. To further stabilize optimization, ICPO integrates Expert Region Reject Sampling to filter unreliable off-policy trajectories and Annealed Expert-Bonus Reward Shaping to balance early expert guidance with later autonomous improvement. Results demonstrate that ICPO consistently enhances reinforcement learning performance and training stability on mathematical reasoning benchmarks, revealing a scalable and effective RLVR paradigm for LRMs.

comment: Work in progress

☆ Inside CORE-KG: Evaluating Structured Prompting and Coreference Resolution for Knowledge Graphs ICDM 2025

Dipak Meher, Carlotta Domeniconi

Human smuggling networks are increasingly adaptive and difficult to analyze. Legal case documents offer critical insights but are often unstructured, lexically dense, and filled with ambiguous or shifting references, which pose significant challenges for automated knowledge graph (KG) construction. While recent LLM-based approaches improve over static templates, they still generate noisy, fragmented graphs with duplicate nodes due to the absence of guided extraction and coreference resolution. The recently proposed CORE-KG framework addresses these limitations by integrating a type-aware coreference module and domain-guided structured prompts, significantly reducing node duplication and legal noise. In this work, we present a systematic ablation study of CORE-KG to quantify the individual contributions of its two key components. Our results show that removing coreference resolution results in a 28.32% increase in node duplication and a 4.32% increase in noisy nodes, while removing structured prompts leads to a 4.34% increase in node duplication and a 73.33% increase in noisy nodes. These findings offer empirical insights for designing robust LLM-based pipelines for extracting structured representations from complex legal texts.

comment: ICDM 2025 Workshop

☆ LLMs as In-Context Meta-Learners for Model and Hyperparameter Selection

Youssef Attia El Hili, Albert Thomas, Malik Tiomoko, Abdelhakim Benechehab, Corentin Léger, Corinne Ancourt, Balázs Kégl

Model and hyperparameter selection are critical but challenging in machine learning, typically requiring expert intuition or expensive automated search. We investigate whether large language models (LLMs) can act as in-context meta-learners for this task. By converting each dataset into interpretable metadata, we prompt an LLM to recommend both model families and hyperparameters. We study two prompting strategies: (1) a zero-shot mode relying solely on pretrained knowledge, and (2) a meta-informed mode augmented with examples of models and their performance on past tasks. Across synthetic and real-world benchmarks, we show that LLMs can exploit dataset metadata to recommend competitive models and hyperparameters without search, and that improvements from meta-informed prompting demonstrate their capacity for in-context meta-learning. These results highlight a promising new role for LLMs as lightweight, general-purpose assistants for model selection and hyperparameter optimization.

comment: 27 pages, 6 figures

☆ Enhancing ECG Classification Robustness with Lightweight Unsupervised Anomaly Detection Filters

Mustafa Fuad Rifet Ibrahim, Maurice Meijer, Alexander Schlaefer, Peer Stelldinger

Continuous electrocardiogram (ECG) monitoring via wearables offers significant potential for early cardiovascular disease (CVD) detection. However, deploying deep learning models for automated analysis in resource-constrained environments faces reliability challenges due to inevitable Out-of-Distribution (OOD) data. OOD inputs, such as unseen pathologies or noisecorrupted signals, often cause erroneous, high-confidence predictions by standard classifiers, compromising patient safety. Existing OOD detection methods either neglect computational constraints or address noise and unseen classes separately. This paper explores Unsupervised Anomaly Detection (UAD) as an independent, upstream filtering mechanism to improve robustness. We benchmark six UAD approaches, including Deep SVDD, reconstruction-based models, Masked Anomaly Detection, normalizing flows, and diffusion models, optimized via Neural Architecture Search (NAS) under strict resource constraints (at most 512k parameters). Evaluation on PTB-XL and BUT QDB datasets assessed detection of OOD CVD classes and signals unsuitable for analysis due to noise. Results show Deep SVDD consistently achieves the best trade-off between detection and efficiency. In a realistic deployment simulation, integrating the optimized Deep SVDD filter with a diagnostic classifier improved accuracy by up to 21 percentage points over a classifier-only baseline. This study demonstrates that optimized UAD filters can safeguard automated ECG analysis, enabling safer, more reliable continuous cardiovascular monitoring on wearables.

comment: Submitted to the 24th International Conference on Pervasive Computing and Communications (PerCom 2026)

☆ Data-Efficient RLVR via Off-Policy Influence Guidance

Erle Zhu, Dazhi Jiang, Yuan Wang, Xujun Li, Jiale Cheng, Yuxian Gu, Yilin Niu, Aohan Zeng, Jie Tang, Minlie Huang, Hongning Wang

Data selection is a critical aspect of Reinforcement Learning with Verifiable Rewards (RLVR) for enhancing the reasoning capabilities of large language models (LLMs). Current data selection methods are largely heuristic-based, lacking theoretical guarantees and generalizability. This work proposes a theoretically-grounded approach using influence functions to estimate the contribution of each data point to the learning objective. To overcome the prohibitive computational cost of policy rollouts required for online influence estimation, we introduce an off-policy influence estimation method that efficiently approximates data influence using pre-collected offline trajectories. Furthermore, to manage the high-dimensional gradients of LLMs, we employ sparse random projection to reduce dimensionality and improve storage and computation efficiency. Leveraging these techniques, we develop \textbf{C}urriculum \textbf{R}L with \textbf{O}ff-\textbf{P}olicy \text{I}nfluence guidance (\textbf{CROPI}), a multi-stage RL framework that iteratively selects the most influential data for the current policy. Experiments on models up to 7B parameters demonstrate that CROPI significantly accelerates training. On a 1.5B model, it achieves a 2.66x step-level acceleration while using only 10\% of the data per stage compared to full-dataset training. Our results highlight the substantial potential of influence-based data selection for efficient RLVR.

☆ Quantum Gated Recurrent GAN with Gaussian Uncertainty for Network Anomaly Detection

Wajdi Hammami, Soumaya Cherkaoui, Jean-Frederic Laprade, Ola Ahmad, Shengrui Wang

Anomaly detection in time-series data is a critical challenge with significant implications for network security. Recent quantum machine learning approaches, such as quantum kernel methods and variational quantum circuits, have shown promise in capturing complex data distributions for anomaly detection but remain constrained by limited qubit counts. We introduce in this work a novel Quantum Gated Recurrent Unit (QGRU)-based Generative Adversarial Network (GAN) employing Successive Data Injection (SuDaI) and a multi-metric gating strategy for robust network anomaly detection. Our model uniquely utilizes a quantum-enhanced generator that outputs parameters (mean and log-variance) of a Gaussian distribution via reparameterization, combined with a Wasserstein critic to stabilize adversarial training. Anomalies are identified through a novel gating mechanism that initially flags potential anomalies based on Gaussian uncertainty estimates and subsequently verifies them using a composite of critic scores and reconstruction errors. Evaluated on benchmark datasets, our method achieves a high time-series aware F1 score (TaF1) of 89.43% demonstrating superior capability in detecting anomalies accurately and promptly as compared to existing classical and quantum models. Furthermore, the trained QGRU-WGAN was deployed on real IBM Quantum hardware, where it retained high anomaly detection performance, confirming its robustness and practical feasibility on current noisy intermediate-scale quantum (NISQ) devices.

☆ LINK-KG: LLM-Driven Coreference-Resolved Knowledge Graphs for Human Smuggling Networks

Dipak Meher, Carlotta Domeniconi, Guadalupe Correa-Cabrera

Human smuggling networks are complex and constantly evolving, making them difficult to analyze comprehensively. Legal case documents offer rich factual and procedural insights into these networks but are often long, unstructured, and filled with ambiguous or shifting references, posing significant challenges for automated knowledge graph (KG) construction. Existing methods either overlook coreference resolution or fail to scale beyond short text spans, leading to fragmented graphs and inconsistent entity linking. We propose LINK-KG, a modular framework that integrates a three-stage, LLM-guided coreference resolution pipeline with downstream KG extraction. At the core of our approach is a type-specific Prompt Cache, which consistently tracks and resolves references across document chunks, enabling clean and disambiguated narratives for structured knowledge graph construction from both short and long legal texts. LINK-KG reduces average node duplication by 45.21% and noisy nodes by 32.22% compared to baseline methods, resulting in cleaner and more coherent graph structures. These improvements establish LINK-KG as a strong foundation for analyzing complex criminal networks.

comment: Accepted in ICKG 2025 Conference, 8 Pages, 2 Figures

☆ ReSpec: Towards Optimizing Speculative Decoding in Reinforcement Learning Systems

Qiaoling Chen, Zijun Liu, Peng Sun, Shenggui Li, Guoteng Wang, Ziming Liu, Yonggang Wen, Siyuan Feng, Tianwei Zhang

Adapting large language models (LLMs) via reinforcement learning (RL) is often bottlenecked by the generation stage, which can consume over 75\% of the training time. Speculative decoding (SD) accelerates autoregressive generation in serving systems, but its behavior under RL training remains largely unexplored. We identify three critical gaps that hinder the naive integration of SD into RL systems: diminishing speedups at large batch sizes, drafter staleness under continual actor updates, and drafter-induced policy degradation. To address these gaps, we present ReSpec, a system that adapts SD to RL through three complementary mechanisms: dynamically tuning SD configurations, evolving the drafter via knowledge distillation, and weighting updates by rollout rewards. On Qwen models (3B--14B), ReSpec achieves up to 4.5x speedup while preserving reward convergence and training stability, providing a practical solution for efficient RL-based LLM adaptation.

☆ Counteracting Matthew Effect in Self-Improvement of LVLMs through Head-Tail Re-balancing

Xin Guo, Zhiheng Xi, Yiwen Ding, Yitao Zhai, Xiaowei Shi, Xunliang Cai, Tao Gui, Qi Zhang, Xuanjing Huang

Self-improvement has emerged as a mainstream paradigm for advancing the reasoning capabilities of large vision-language models (LVLMs), where models explore and learn from successful trajectories iteratively. However, we identify a critical issue during this process: the model excels at generating high-quality trajectories for simple queries (i.e., head data) but struggles with more complex ones (i.e., tail data). This leads to an imbalanced optimization that drives the model to prioritize simple reasoning skills, while hindering its ability to tackle more complex reasoning tasks. Over iterations, this imbalance becomes increasingly pronounced--a dynamic we term the "Matthew effect"--which ultimately hinders further model improvement and leads to performance bottlenecks. To counteract this challenge, we introduce four efficient strategies from two perspectives: distribution-reshaping and trajectory-resampling, to achieve head-tail re-balancing during the exploration-and-learning self-improvement process. Extensive experiments on Qwen2-VL-7B-Instruct and InternVL2.5-4B models across visual reasoning tasks demonstrate that our methods consistently improve visual reasoning capabilities, outperforming vanilla self-improvement by 3.86 points on average.

comment: Preprint

☆ Representation-Level Counterfactual Calibration for Debiased Zero-Shot Recognition

Pei Peng, MingKun Xie, Hang Hao, Tong Jin, ShengJun Huang

Object-context shortcuts remain a persistent challenge in vision-language models, undermining zero-shot reliability when test-time scenes differ from familiar training co-occurrences. We recast this issue as a causal inference problem and ask: Would the prediction remain if the object appeared in a different environment? To answer this at inference time, we estimate object and background expectations within CLIP's representation space, and synthesize counterfactual embeddings by recombining object features with diverse alternative contexts sampled from external datasets, batch neighbors, or text-derived descriptions. By estimating the Total Direct Effect and simulating intervention, we further subtract background-only activation, preserving beneficial object-context interactions while mitigating hallucinated scores. Without retraining or prompt design, our method substantially improves both worst-group and average accuracy on context-sensitive benchmarks, establishing a new zero-shot state of the art. Beyond performance, our framework provides a lightweight representation-level counterfactual approach, offering a practical causal avenue for debiased and reliable multimodal reasoning.

☆ Vectorized Context-Aware Embeddings for GAT-Based Collaborative Filtering

Danial Ebrat, Sepideh Ahmadian, Luis Rueda

Recommender systems often struggle with data sparsity and cold-start scenarios, limiting their ability to provide accurate suggestions for new or infrequent users. This paper presents a Graph Attention Network (GAT) based Collaborative Filtering (CF) framework enhanced with Large Language Model (LLM) driven context aware embeddings. Specifically, we generate concise textual user profiles and unify item metadata (titles, genres, overviews) into rich textual embeddings, injecting these as initial node features in a bipartite user item graph. To further optimize ranking performance, we introduce a hybrid loss function that combines Bayesian Personalized Ranking (BPR) with a cosine similarity term and robust negative sampling, ensuring explicit negative feedback is distinguished from unobserved data. Experiments on the MovieLens 100k and 1M datasets show consistent improvements over state-of-the-art baselines in Precision, NDCG, and MAP while demonstrating robustness for users with limited interaction history. Ablation studies confirm the critical role of LLM-augmented embeddings and the cosine similarity term in capturing nuanced semantic relationships. Our approach effectively mitigates sparsity and cold-start limitations by integrating LLM-derived contextual understanding into graph-based architectures. Future directions include balancing recommendation accuracy with coverage and diversity, and introducing fairness-aware constraints and interpretability features to enhance system performance further.

☆ Robust Graph Condensation via Classification Complexity Mitigation

Jiayi Luo, Qingyun Sun, Beining Yang, Haonan Yuan, Xingcheng Fu, Yanbiao Ma, Jianxin Li, Philip S. Yu

Graph condensation (GC) has gained significant attention for its ability to synthesize smaller yet informative graphs. However, existing studies often overlook the robustness of GC in scenarios where the original graph is corrupted. In such cases, we observe that the performance of GC deteriorates significantly, while existing robust graph learning technologies offer only limited effectiveness. Through both empirical investigation and theoretical analysis, we reveal that GC is inherently an intrinsic-dimension-reducing process, synthesizing a condensed graph with lower classification complexity. Although this property is critical for effective GC performance, it remains highly vulnerable to adversarial perturbations. To tackle this vulnerability and improve GC robustness, we adopt the geometry perspective of graph data manifold and propose a novel Manifold-constrained Robust Graph Condensation framework named MRGC. Specifically, we introduce three graph data manifold learning modules that guide the condensed graph to lie within a smooth, low-dimensional manifold with minimal class ambiguity, thereby preserving the classification complexity reduction capability of GC and ensuring robust performance under universal adversarial attacks. Extensive experiments demonstrate the robustness of \ModelName\ across diverse attack scenarios.

☆ Personalized Treatment Outcome Prediction from Scarce Data via Dual-Channel Knowledge Distillation and Adaptive Fusion

Wenjie Chen, Li Zhuang, Ziying Luo, Yu Liu, Jiahao Wu, Shengcai Liu

Personalized treatment outcome prediction based on trial data for small-sample and rare patient groups is critical in precision medicine. However, the costly trial data limit the prediction performance. To address this issue, we propose a cross-fidelity knowledge distillation and adaptive fusion network (CFKD-AFN), which leverages abundant but low-fidelity simulation data to enhance predictions on scarce but high-fidelity trial data. CFKD-AFN incorporates a dual-channel knowledge distillation module to extract complementary knowledge from the low-fidelity model, along with an attention-guided fusion module to dynamically integrate multi-source information. Experiments on treatment outcome prediction for the chronic obstructive pulmonary disease demonstrates significant improvements of CFKD-AFN over state-of-the-art methods in prediction accuracy, ranging from 6.67\% to 74.55\%, and strong robustness to varying high-fidelity dataset sizes. Furthermore, we extend CFKD-AFN to an interpretable variant, enabling the exploration of latent medical semantics to support clinical decision-making.

☆ Co-Evolving Latent Action World Models

Yucen Wang, Fengming Zhang, De-Chuan Zhan, Li Zhao, Kaixin Wang, Jiang Bian

Adapting pre-trained video generation models into controllable world models via latent actions is a promising step towards creating generalist world models. The dominant paradigm adopts a two-stage approach that trains latent action model (LAM) and the world model separately, resulting in redundant training and limiting their potential for co-adaptation. A conceptually simple and appealing idea is to directly replace the forward dynamic model in LAM with a powerful world model and training them jointly, but it is non-trivial and prone to representational collapse. In this work, we propose CoLA-World, which for the first time successfully realizes this synergistic paradigm, resolving the core challenge in joint learning through a critical warm-up phase that effectively aligns the representations of the from-scratch LAM with the pre-trained world model. This unlocks a co-evolution cycle: the world model acts as a knowledgeable tutor, providing gradients to shape a high-quality LAM, while the LAM offers a more precise and adaptable control interface to the world model. Empirically, CoLA-World matches or outperforms prior two-stage methods in both video simulation quality and downstream visual planning, establishing a robust and efficient new paradigm for the field.

☆ Autograder+: A Multi-Faceted AI Framework for Rich Pedagogical Feedback in Programming Education

Vikrant Sahu, Gagan Raj Gupta, Raghav Borikar, Nitin Mane

The rapid growth of programming education has outpaced traditional assessment tools, leaving faculty with limited means to provide meaningful, scalable feedback. Conventional autograders, while efficient, act as black-box systems that simply return pass/fail results, offering little insight into student thinking or learning needs. Autograder+ is designed to shift autograding from a purely summative process to a formative learning experience. It introduces two key capabilities: automated feedback generation using a fine-tuned Large Language Model, and visualization of student code submissions to uncover learning patterns. The model is fine-tuned on curated student code and expert feedback to ensure pedagogically aligned, context-aware guidance. In evaluation across 600 student submissions from multiple programming tasks, the system produced feedback with strong semantic alignment to instructor comments. For visualization, contrastively learned code embeddings trained on 1,000 annotated submissions enable grouping solutions into meaningful clusters based on functionality and approach. The system also supports prompt-pooling, allowing instructors to guide feedback style through selected prompt templates. By integrating AI-driven feedback, semantic clustering, and interactive visualization, Autograder+ reduces instructor workload while supporting targeted instruction and promoting stronger learning outcomes.

☆ Multi-Output Robust and Conjugate Gaussian Processes

Joshua Rooijakkers, Leiv Rønneberg, François-Xavier Briol, Jeremias Knoblauch, Matias Altamirano

Multi-output Gaussian process (MOGP) regression allows modelling dependencies among multiple correlated response variables. Similarly to standard Gaussian processes, MOGPs are sensitive to model misspecification and outliers, which can distort predictions within individual outputs. This situation can be further exacerbated by multiple anomalous response variables whose errors propagate due to correlations between outputs. To handle this situation, we extend and generalise the robust and conjugate Gaussian process (RCGP) framework introduced by Altamirano et al. (2024). This results in the multi-output RCGP (MO-RCGP): a provably robust MOGP that is conjugate, and jointly captures correlations across outputs. We thoroughly evaluate our approach through applications in finance and cancer research.

☆ Multi-Task Learning Based on Support Vector Machines and Twin Support Vector Machines: A Comprehensive Survey

Fatemeh Bazikar, Hossein Moosaei, Atefeh Hemmati, Panos M. Pardalos

Multi-task learning (MTL) enables simultaneous training across related tasks, leveraging shared information to improve generalization, efficiency, and robustness, especially in data-scarce or high-dimensional scenarios. While deep learning dominates recent MTL research, Support Vector Machines (SVMs) and Twin SVMs (TWSVMs) remain relevant due to their interpretability, theoretical rigor, and effectiveness with small datasets. This chapter surveys MTL approaches based on SVM and TWSVM, highlighting shared representations, task regularization, and structural coupling strategies. Special attention is given to emerging TWSVM extensions for multi-task settings, which show promise but remain underexplored. We compare these models in terms of theoretical properties, optimization strategies, and empirical performance, and discuss applications in fields such as computer vision, natural language processing, and bioinformatics. Finally, we identify research gaps and outline future directions for building scalable, interpretable, and reliable margin-based MTL frameworks. This work provides a comprehensive resource for researchers and practitioners interested in SVM- and TWSVM-based multi-task learning.

☆ Adaptive Context Length Optimization with Low-Frequency Truncation for Multi-Agent Reinforcement Learning

Wenchang Duan, Yaoliang Yu, Jiwan He, Yi Shi

Recently, deep multi-agent reinforcement learning (MARL) has demonstrated promising performance for solving challenging tasks, such as long-term dependencies and non-Markovian environments. Its success is partly attributed to conditioning policies on large fixed context length. However, such large fixed context lengths may lead to limited exploration efficiency and redundant information. In this paper, we propose a novel MARL framework to obtain adaptive and effective contextual information. Specifically, we design a central agent that dynamically optimizes context length via temporal gradient analysis, enhancing exploration to facilitate convergence to global optima in MARL. Furthermore, to enhance the adaptive optimization capability of the context length, we present an efficient input representation for the central agent, which effectively filters redundant information. By leveraging a Fourier-based low-frequency truncation method, we extract global temporal trends across decentralized agents, providing an effective and efficient representation of the MARL environment. Extensive experiments demonstrate that the proposed method achieves state-of-the-art (SOTA) performance on long-term dependency tasks, including PettingZoo, MiniGrid, Google Research Football (GRF), and StarCraft Multi-Agent Challenge v2 (SMACv2).

☆ Scales++: Compute Efficient Evaluation Subset Selection with Cognitive Scales Embeddings

Andrew M. Bean, Nabeel Seedat, Shengzhuang Chen, Jonathan Richard Schwarz

The prohibitive cost of evaluating large language models (LLMs) on comprehensive benchmarks necessitates the creation of small yet representative data subsets (i.e., tiny benchmarks) that enable efficient assessment while retaining predictive fidelity. Current methods for this task operate under a model-centric paradigm, selecting benchmarking items based on the collective performance of existing models. Such approaches are limited by large upfront costs, an inability to immediately handle new benchmarks (`cold-start'), and the fragile assumption that future models will share the failure patterns of their predecessors. In this work, we challenge this paradigm and propose a item-centric approach to benchmark subset selection, arguing that selection should be based on the intrinsic properties of the task items themselves, rather than on model-specific failure patterns. We instantiate this item-centric efficient benchmarking approach via a novel method, Scales++, where data selection is based on the cognitive demands of the benchmark samples. Empirically, we show Scales++ reduces the upfront selection cost by over 18x while achieving competitive predictive fidelity. On the Open LLM Leaderboard, using just a 0.5\% data subset, we predict full benchmark scores with a 2.9% mean absolute error. We demonstrate that this item-centric approach enables more efficient model evaluation without significant fidelity degradation, while also providing better cold-start performance and more interpretable benchmarking.

comment: 9 pages, 2 figures, 4 tables

☆ Efficient Generative AI Boosts Probabilistic Forecasting of Sudden Stratospheric Warmings

Ningning Tao, Fei Xie, Baoxiang Pan, Hongyu Wang, Han Huang, Zhongpu Qiu, Ke Gui, Jiali Luo, Xiaosong Chen

Sudden Stratospheric Warmings (SSWs) are key sources of subseasonal predictability and major drivers of extreme winter weather. Yet, their accurate and efficient forecast remains a persistent challenge for numerical weather prediction (NWP) systems due to limitations in physical representation, initialization, and the immense computational demands of ensemble forecasts. While data-driven forecasting is rapidly evolving, its application to the complex, three-dimensional dynamics of SSWs, particularly for probabilistic forecast, remains underexplored. Here, we bridge this gap by developing a Flow Matching-based generative AI model (FM-Cast) for efficient and skillful probabilistic forecasting of the spatiotemporal evolution of stratospheric circulation. Evaluated across 18 major SSW events (1998-2024), FM-Cast skillfully forecasts the onset, intensity, and morphology of 10 events up to 20 days in advance, achieving ensemble accuracies above 50%. Its performance is comparable to or exceeds leading NWP systems while requiring only two minutes for a 50-member, 30-day forecast on a consumer GPU. Furthermore, leveraging FM-Cast as a scientific tool, we demonstrate through idealized experiments that SSW predictability is fundamentally linked to its underlying physical drivers, distinguishing between events forced from the troposphere and those driven by internal stratospheric dynamics. Our work thus establishes a computationally efficient paradigm for probabilistic forecasting stratospheric anomalies and showcases generative AI's potential to deepen the physical understanding of atmosphere-climate dynamics.

☆ CorVS: Person Identification via Video Trajectory-Sensor Correspondence in a Real-World Warehouse

Kazuma Kano, Yuki Mori, Shin Katayama, Kenta Urano, Takuro Yonezawa, Nobuo Kawaguchi

Worker location data is key to higher productivity in industrial sites. Cameras are a promising tool for localization in logistics warehouses since they also offer valuable environmental contexts such as package status. However, identifying individuals with only visual data is often impractical. Accordingly, several prior studies identified people in videos by comparing their trajectories and wearable sensor measurements. While this approach has advantages such as independence from appearance, the existing methods may break down under real-world conditions. To overcome this challenge, we propose CorVS, a novel data-driven person identification method based on correspondence between visual tracking trajectories and sensor measurements. Firstly, our deep learning model predicts correspondence probabilities and reliabilities for every pair of a trajectory and sensor measurements. Secondly, our algorithm matches the trajectories and sensor measurements over time using the predicted probabilities and reliabilities. We developed a dataset with actual warehouse operations and demonstrated the method's effectiveness for real-world applications.

comment: 7 pages, 3 figures, accepted to IPIN 2025

☆ Towards Explainable and Reliable AI in Finance

Albi Isufaj, Pablo Mollá, Helmut Prendinger

Financial forecasting increasingly uses large neural network models, but their opacity raises challenges for trust and regulatory compliance. We present several approaches to explainable and reliable AI in finance. \emph{First}, we describe how Time-LLM, a time series foundation model, uses a prompt to avoid a wrong directional forecast. \emph{Second}, we show that combining foundation models for time series forecasting with a reliability estimator can filter our unreliable predictions. \emph{Third}, we argue for symbolic reasoning encoding domain rules for transparent justification. These approaches shift emphasize executing only forecasts that are both reliable and explainable. Experiments on equity and cryptocurrency data show that the architecture reduces false positives and supports selective execution. By integrating predictive performance with reliability estimation and rule-based reasoning, our framework advances transparent and auditable financial AI systems.

☆ UnifiedFL: A Dynamic Unified Learning Framework for Equitable Federation

Furkan Pala, Islem Rekik

Federated learning (FL) has emerged as a key paradigm for collaborative model training across multiple clients without sharing raw data, enabling privacy-preserving applications in areas such as radiology and pathology. However, works on collaborative training across clients with fundamentally different neural architectures and non-identically distributed datasets remain scarce. Existing FL frameworks face several limitations. Despite claiming to support architectural heterogeneity, most recent FL methods only tolerate variants within a single model family (e.g., shallower, deeper, or wider CNNs), still presuming a shared global architecture and failing to accommodate federations where clients deploy fundamentally different network types (e.g., CNNs, GNNs, MLPs). Moreover, existing approaches often address only statistical heterogeneity while overlooking the domain-fracture problem, where each client's data distribution differs markedly from that faced at testing time, undermining model generalizability. When clients use different architectures, have non-identically distributed data, and encounter distinct test domains, current methods perform poorly. To address these challenges, we propose UnifiedFL, a dynamic federated learning framework that represents heterogeneous local networks as nodes and edges in a directed model graph optimized by a shared graph neural network (GNN). UnifiedFL introduces (i) a common GNN to parameterize all architectures, (ii) distance-driven clustering via Euclidean distances between clients' parameters, and (iii) a two-tier aggregation policy balancing convergence and diversity. Experiments on MedMNIST classification and hippocampus segmentation benchmarks demonstrate UnifiedFL's superior performance. Code and data: https://github.com/basiralab/UnifiedFL

☆ Reinforcement Learning for Pollution Detection in a Randomized, Sparse and Nonstationary Environment with an Autonomous Underwater Vehicle

Sebastian Zieglmeier, Niklas Erdmann, Narada D. Warakagoda

Reinforcement learning (RL) algorithms are designed to optimize problem-solving by learning actions that maximize rewards, a task that becomes particularly challenging in random and nonstationary environments. Even advanced RL algorithms are often limited in their ability to solve problems in these conditions. In applications such as searching for underwater pollution clouds with autonomous underwater vehicles (AUVs), RL algorithms must navigate reward-sparse environments, where actions frequently result in a zero reward. This paper aims to address these challenges by revisiting and modifying classical RL approaches to efficiently operate in sparse, randomized, and nonstationary environments. We systematically study a large number of modifications, including hierarchical algorithm changes, multigoal learning, and the integration of a location memory as an external output filter to prevent state revisits. Our results demonstrate that a modified Monte Carlo-based approach significantly outperforms traditional Q-learning and two exhaustive search patterns, illustrating its potential in adapting RL to complex environments. These findings suggest that reinforcement learning approaches can be effectively adapted for use in random, nonstationary, and reward-sparse environments.

☆ MisSynth: Improving MISSCI Logical Fallacies Classification with Synthetic Data

Mykhailo Poliakov, Nadiya Shvai

Health-related misinformation is very prevalent and potentially harmful. It is difficult to identify, especially when claims distort or misinterpret scientific findings. We investigate the impact of synthetic data generation and lightweight fine-tuning techniques on the ability of large language models (LLMs) to recognize fallacious arguments using the MISSCI dataset and framework. In this work, we propose MisSynth, a pipeline that applies retrieval-augmented generation (RAG) to produce synthetic fallacy samples, which are then used to fine-tune an LLM model. Our results show substantial accuracy gains with fine-tuned models compared to vanilla baselines. For instance, the LLaMA 3.1 8B fine-tuned model achieved an over 35% F1-score absolute improvement on the MISSCI test split over its vanilla baseline. We demonstrate that introducing synthetic fallacy data to augment limited annotated resources can significantly enhance zero-shot LLM classification performance on real-world scientific misinformation tasks, even with limited computational resources. The code and synthetic dataset are available on https://github.com/mxpoliakov/MisSynth.

☆ Linear Causal Discovery with Interventional Constraints

Zhigao Guo, Feng Dong

Incorporating causal knowledge and mechanisms is essential for refining causal models and improving downstream tasks such as designing new treatments. In this paper, we introduce a novel concept in causal discovery, termed interventional constraints, which differs fundamentally from interventional data. While interventional data require direct perturbations of variables, interventional constraints encode high-level causal knowledge in the form of inequality constraints on causal effects. For instance, in the Sachs dataset (Sachs et al.\ 2005), Akt has been shown to be activated by PIP3, meaning PIP3 exerts a positive causal effect on Akt. Existing causal discovery methods allow enforcing structural constraints (for example, requiring a causal path from PIP3 to Akt), but they may still produce incorrect causal conclusions such as learning that "PIP3 inhibits Akt". Interventional constraints bridge this gap by explicitly constraining the total causal effect between variable pairs, ensuring learned models respect known causal influences. To formalize interventional constraints, we propose a metric to quantify total causal effects for linear causal models and formulate the problem as a constrained optimization task, solved using a two-stage constrained optimization method. We evaluate our approach on real-world datasets and demonstrate that integrating interventional constraints not only improves model accuracy and ensures consistency with established findings, making models more explainable, but also facilitates the discovery of new causal relationships that would otherwise be costly to identify.

☆ SABER: Symbolic Regression-based Angle of Arrival and Beam Pattern Estimator

Shih-Kai Chou, Mengran Zhao, Cheng-Nan Hu, Kuang-Chung Chou, Carolina Fortuna, Jernej Hribar

Accurate Angle-of-arrival (AoA) estimation is essential for next-generation wireless communication systems to enable reliable beamforming, high-precision localization, and integrated sensing. Unfortunately, classical high-resolution techniques require multi-element arrays and extensive snapshot collection, while generic Machine Learning (ML) approaches often yield black-box models that lack physical interpretability. To address these limitations, we propose a Symbolic Regression (SR)-based ML framework. Namely, Symbolic Regression-based Angle of Arrival and Beam Pattern Estimator (SABER), a constrained symbolic-regression framework that automatically discovers closed-form beam pattern and AoA models from path loss measurements with interpretability. SABER achieves high accuracy while bridging the gap between opaque ML methods and interpretable physics-driven estimators. First, we validate our approach in a controlled free-space anechoic chamber, showing that both direct inversion of the known $\cos^n$ beam and a low-order polynomial surrogate achieve sub-0.5 degree Mean Absolute Error (MAE). A purely unconstrained SR method can further reduce the error of the predicted angles, but produces complex formulas that lack physical insight. Then, we implement the same SR-learned inversions in a real-world, Reconfigurable Intelligent Surface (RIS)-aided indoor testbed. SABER and unconstrained SR models accurately recover the true AoA with near-zero error. Finally, we benchmark SABER against the Cram\'er-Rao Lower Bounds (CRLBs). Our results demonstrate that SABER is an interpretable and accurate alternative to state-of-the-art and black-box ML-based methods for AoA estimation.

comment: 12 pages, 11 figures

☆ Agent Skills Enable a New Class of Realistic and Trivially Simple Prompt Injections

David Schmotz, Sahar Abdelnabi, Maksym Andriushchenko

Enabling continual learning in LLMs remains a key unresolved research challenge. In a recent announcement, a frontier LLM company made a step towards this by introducing Agent Skills, a framework that equips agents with new knowledge based on instructions stored in simple markdown files. Although Agent Skills can be a very useful tool, we show that they are fundamentally insecure, since they enable trivially simple prompt injections. We demonstrate how to hide malicious instructions in long Agent Skill files and referenced scripts to exfiltrate sensitive data, such as internal files or passwords. Importantly, we show how to bypass system-level guardrails of a popular coding agent: a benign, task-specific approval with the "Don't ask again" option can carry over to closely related but harmful actions. Overall, we conclude that despite ongoing research efforts and scaling model capabilities, frontier LLMs remain vulnerable to very simple prompt injections in realistic scenarios. Our code is available at https://github.com/aisa-group/promptinject-agent-skills.

☆ Posterior Sampling by Combining Diffusion Models with Annealed Langevin Dynamics NeurIPS 2025

Zhiyang Xun, Shivam Gupta, Eric Price

Given a noisy linear measurement $y = Ax + \xi$ of a distribution $p(x)$, and a good approximation to the prior $p(x)$, when can we sample from the posterior $p(x \mid y)$? Posterior sampling provides an accurate and fair framework for tasks such as inpainting, deblurring, and MRI reconstruction, and several heuristics attempt to approximate it. Unfortunately, approximate posterior sampling is computationally intractable in general. To sidestep this hardness, we focus on (local or global) log-concave distributions $p(x)$. In this regime, Langevin dynamics yields posterior samples when the exact scores of $p(x)$ are available, but it is brittle to score--estimation error, requiring an MGF bound (sub-exponential error). By contrast, in the unconditional setting, diffusion models succeed with only an $L^2$ bound on the score error. We prove that combining diffusion models with an annealed variant of Langevin dynamics achieves conditional sampling in polynomial time using merely an $L^4$ bound on the score error.

comment: NeurIPS 2025

☆ On the Impact of Weight Discretization in QUBO-Based SVM Training ECML

Sascha Mücke

Training Support Vector Machines (SVMs) can be formulated as a QUBO problem, enabling the use of quantum annealing for model optimization. In this work, we study how the number of qubits - linked to the discretization level of dual weights - affects predictive performance across datasets. We compare QUBO-based SVM training to the classical LIBSVM solver and find that even low-precision QUBO encodings (e.g., 1 bit per parameter) yield competitive, and sometimes superior, accuracy. While increased bit-depth enables larger regularization parameters, it does not always improve classification. Our findings suggest that selecting the right support vectors may matter more than their precise weighting. Although current hardware limits the size of solvable QUBOs, our results highlight the potential of quantum annealing for efficient SVM training as quantum devices scale.

comment: Presented at the 7th DSO Workshop at ECML PKDD 2025

☆ Model Inversion with Layer-Specific Modeling and Alignment for Data-Free Continual Learning NeurIPS 2025

Ruilin Tong, Haodong Lu, Yuhang Liu, Dong Gong

Continual learning (CL) aims to incrementally train a model on a sequence of tasks while retaining performance on prior ones. However, storing and replaying data is often infeasible due to privacy or security constraints and impractical for arbitrary pre-trained models. Data-free CL seeks to update models without access to previous data. Beyond regularization, we employ model inversion to synthesize data from the trained model, enabling replay without storing samples. Yet, model inversion in predictive models faces two challenges: (1) generating inputs solely from compressed output labels causes drift between synthetic and real data, and replaying such data can erode prior knowledge; (2) inversion is computationally expensive since each step backpropagates through the full model. These issues are amplified in large pre-trained models such as CLIP. To improve efficiency, we propose Per-layer Model Inversion (PMI), inspired by faster convergence in single-layer optimization. PMI provides strong initialization for full-model inversion, substantially reducing iterations. To mitigate feature shift, we model class-wise features via Gaussian distributions and contrastive model, ensuring alignment between synthetic and real features. Combining PMI and feature modeling, our approach enables continual learning of new classes by generating pseudo-images from semantic-aware projected features, achieving strong effectiveness and compatibility across multiple CL settings.

comment: Accepted in NeurIPS 2025

☆ A Survey of Heterogeneous Graph Neural Networks for Cybersecurity Anomaly Detection

Laura Jiang, Reza Ryan, Qian Li, Nasim Ferdosian

Anomaly detection is a critical task in cybersecurity, where identifying insider threats, access violations, and coordinated attacks is essential for ensuring system resilience. Graph-based approaches have become increasingly important for modeling entity interactions, yet most rely on homogeneous and static structures, which limits their ability to capture the heterogeneity and temporal evolution of real-world environments. Heterogeneous Graph Neural Networks (HGNNs) have emerged as a promising paradigm for anomaly detection by incorporating type-aware transformations and relation-sensitive aggregation, enabling more expressive modeling of complex cyber data. However, current research on HGNN-based anomaly detection remains fragmented, with diverse modeling strategies, limited comparative evaluation, and an absence of standardized benchmarks. To address this gap, we provide a comprehensive survey of HGNN-based anomaly detection methods in cybersecurity. We introduce a taxonomy that classifies approaches by anomaly type and graph dynamics, analyze representative models, and map them to key cybersecurity applications. We also review commonly used benchmark datasets and evaluation metrics, highlighting their strengths and limitations. Finally, we identify key open challenges related to modeling, data, and deployment, and outline promising directions for future research. This survey aims to establish a structured foundation for advancing HGNN-based anomaly detection toward scalable, interpretable, and practically deployable solutions.

comment: 37 pages, 4 figures, 86 references. Submitted to Journal of Computer Security (under review)

☆ Implicit Bias of Per-sample Adam on Separable Data: Departure from the Full-batch Regime

Beomhan Baek, Minhak Song, Chulhee Yun

Adam [Kingma and Ba, 2015] is the de facto optimizer in deep learning, yet its theoretical understanding remains limited. Prior analyses show that Adam favors solutions aligned with $\ell_\infty$-geometry, but these results are restricted to the full-batch regime. In this work, we study the implicit bias of incremental Adam (using one sample per step) for logistic regression on linearly separable data, and we show that its bias can deviate from the full-batch behavior. To illustrate this, we construct a class of structured datasets where incremental Adam provably converges to the $\ell_2$-max-margin classifier, in contrast to the $\ell_\infty$-max-margin bias of full-batch Adam. For general datasets, we develop a proxy algorithm that captures the limiting behavior of incremental Adam as $\beta_2 \to 1$ and we characterize its convergence direction via a data-dependent dual fixed-point formulation. Finally, we prove that, unlike Adam, Signum [Bernstein et al., 2018] converges to the $\ell_\infty$-max-margin classifier for any batch size by taking $\beta$ close enough to 1. Overall, our results highlight that the implicit bias of Adam crucially depends on both the batching scheme and the dataset, while Signum remains invariant.

comment: 50 pages

☆ Understanding Hardness of Vision-Language Compositionality from A Token-level Causal Lens

Ziliang Chen, Tianang Xiao, Jusheng Zhang, Yongsen Zheng, Xipeng Chen

Contrastive Language-Image Pre-training (CLIP) delivers strong cross modal generalization by aligning images and texts in a shared embedding space, yet it persistently fails at compositional reasoning over objects, attributes, and relations often behaving like a bag-of-words matcher. Prior causal accounts typically model text as a single vector, obscuring token-level structure and leaving core phenomena-such as prompt sensitivity and failures on hard negatives unexplained. We address this gap with a token-aware causal representation learning (CRL) framework grounded in a sequential, language-token SCM. Our theory extends block identifiability to tokenized text, proving that CLIP's contrastive objective can recover the modal-invariant latent variable under both sentence-level and token-level SCMs. Crucially, token granularity yields the first principled explanation of CLIP's compositional brittleness: composition nonidentifiability. We show the existence of pseudo-optimal text encoders that achieve perfect modal-invariant alignment yet are provably insensitive to SWAP, REPLACE, and ADD operations over atomic concepts, thereby failing to distinguish correct captions from hard negatives despite optimizing the same training objective as true-optimal encoders. The analysis further links language-side nonidentifiability to visual-side failures via the modality gap and shows how iterated composition operators compound hardness, motivating improved negative mining strategies.

☆ Offline Clustering of Preference Learning with Active-data Augmentation

Jingyuan Liu, Fatemeh Ghaffari, Xuchuang Wang, Mohammad Hajiesmaili, Carlee Joe-Wong

Preference learning from pairwise feedback is a widely adopted framework in applications such as reinforcement learning with human feedback and recommendations. In many practical settings, however, user interactions are limited or costly, making offline preference learning necessary. Moreover, real-world preference learning often involves users with different preferences. For example, annotators from different backgrounds may rank the same responses differently. This setting presents two central challenges: (1) identifying similarity across users to effectively aggregate data, especially under scenarios where offline data is imbalanced across dimensions, and (2) handling the imbalanced offline data where some preference dimensions are underrepresented. To address these challenges, we study the Offline Clustering of Preference Learning problem, where the learner has access to fixed datasets from multiple users with potentially different preferences and aims to maximize utility for a test user. To tackle the first challenge, we first propose Off-C$^2$PL for the pure offline setting, where the learner relies solely on offline data. Our theoretical analysis provides a suboptimality bound that explicitly captures the tradeoff between sample noise and bias. To address the second challenge of inbalanced data, we extend our framework to the setting with active-data augmentation where the learner is allowed to select a limited number of additional active-data for the test user based on the cluster structure learned by Off-C$^2$PL. In this setting, our second algorithm, A$^2$-Off-C$^2$PL, actively selects samples that target the least-informative dimensions of the test user's preference. We prove that these actively collected samples contribute more effectively than offline ones. Finally, we validate our theoretical results through simulations on synthetic and real-world datasets.

☆ Unravelling the Mechanisms of Manipulating Numbers in Language Models

Michal Štefánik, Timothee Mickus, Marek Kadlčík, Bertram Højer, Michal Spiegel, Raúl Vázquez, Aman Sinha, Josef Kuchař, Philipp Mondorf

Recent work has shown that different large language models (LLMs) converge to similar and accurate input embedding representations for numbers. These findings conflict with the documented propensity of LLMs to produce erroneous outputs when dealing with numeric information. In this work, we aim to explain this conflict by exploring how language models manipulate numbers and quantify the lower bounds of accuracy of these mechanisms. We find that despite surfacing errors, different language models learn interchangeable representations of numbers that are systematic, highly accurate and universal across their hidden states and the types of input contexts. This allows us to create universal probes for each LLM and to trace information -- including the causes of output errors -- to specific layers. Our results lay a fundamental understanding of how pre-trained LLMs manipulate numbers and outline the potential of more accurate probing techniques in addressed refinements of LLMs' architectures.

☆ Empirical Bayesian Multi-Bandit Learning

Xia Jiang, Rong J. B. Zhu

Multi-task learning in contextual bandits has attracted significant research interest due to its potential to enhance decision-making across multiple related tasks by leveraging shared structures and task-specific heterogeneity. In this article, we propose a novel hierarchical Bayesian framework for learning in various bandit instances. This framework captures both the heterogeneity and the correlations among different bandit instances through a hierarchical Bayesian model, enabling effective information sharing while accommodating instance-specific variations. Unlike previous methods that overlook the learning of the covariance structure across bandits, we introduce an empirical Bayesian approach to estimate the covariance matrix of the prior distribution.This enhances both the practicality and flexibility of learning across multi-bandits. Building on this approach, we develop two efficient algorithms: ebmTS (Empirical Bayesian Multi-Bandit Thompson Sampling) and ebmUCB (Empirical Bayesian Multi-Bandit Upper Confidence Bound), both of which incorporate the estimated prior into the decision-making process. We provide the frequentist regret upper bounds for the proposed algorithms, thereby filling a research gap in the field of multi-bandit problems. Extensive experiments on both synthetic and real-world datasets demonstrate the superior performance of our algorithms, particularly in complex environments. Our methods achieve lower cumulative regret compared to existing techniques, highlighting their effectiveness in balancing exploration and exploitation across multi-bandits.

comment: 33 pages, 13 figures

☆ Distributional Multi-objective Black-box Optimization for Diffusion-model Inference-time Multi-Target Generation

Kim Yong Tan, Yueming Lyu, Ivor Tsang, Yew-Soon Ong

Diffusion models have been successful in learning complex data distributions. This capability has driven their application to high-dimensional multi-objective black-box optimization problem. Existing approaches often employ an external optimization loop, such as an evolutionary algorithm, to the diffusion model. However, these approaches treat the diffusion model as a black-box refiner, which overlooks the internal distribution transition of the diffusion generation process, limiting their efficiency. To address these challenges, we propose the Inference-time Multi-target Generation (IMG) algorithm, which optimizes the diffusion process at inference-time to generate samples that simultaneously satisfy multiple objectives. Specifically, our IMG performs weighted resampling during the diffusion generation process according to the expected aggregated multi-objective values. This weighted resampling strategy ensures the diffusion-generated samples are distributed according to our desired multi-target Boltzmann distribution. We further derive that the multi-target Boltzmann distribution has an interesting log-likelihood interpretation, where it is the optimal solution to the distributional multi-objective optimization problem. We implemented IMG for a multi-objective molecule generation task. Experiments show that IMG, requiring only a single generation pass, achieves a significantly higher hypervolume than baseline optimization algorithms that often require hundreds of diffusion generations. Notably, our algorithm can be viewed as an optimized diffusion process and can be integrated into existing methods to further improve their performance.

☆ A Research Roadmap for Augmenting Software Engineering Processes and Software Products with Generative AI

Domenico Amalfitano, Andreas Metzger, Marco Autili, Tommaso Fulcini, Tobias Hey, Jan Keim, Patrizio Pelliccione, Vincenzo Scotti, Anne Koziolek, Raffaela Mirandola, Andreas Vogelsang

Generative AI (GenAI) is rapidly transforming software engineering (SE) practices, influencing how SE processes are executed, as well as how software systems are developed, operated, and evolved. This paper applies design science research to build a roadmap for GenAI-augmented SE. The process consists of three cycles that incrementally integrate multiple sources of evidence, including collaborative discussions from the FSE 2025 "Software Engineering 2030" workshop, rapid literature reviews, and external feedback sessions involving peers. McLuhan's tetrads were used as a conceptual instrument to systematically capture the transforming effects of GenAI on SE processes and software products.The resulting roadmap identifies four fundamental forms of GenAI augmentation in SE and systematically characterizes their related research challenges and opportunities. These insights are then consolidated into a set of future research directions. By grounding the roadmap in a rigorous multi-cycle process and cross-validating it among independent author teams and peers, the study provides a transparent and reproducible foundation for analyzing how GenAI affects SE processes, methods and tools, and for framing future research within this rapidly evolving area. Based on these findings, the article finally makes ten predictions for SE in the year 2030.

☆ PVMark: Enabling Public Verifiability for LLM Watermarking Schemes

Haohua Duan, Liyao Xiang, Xin Zhang

Watermarking schemes for large language models (LLMs) have been proposed to identify the source of the generated text, mitigating the potential threats emerged from model theft. However, current watermarking solutions hardly resolve the trust issue: the non-public watermark detection cannot prove itself faithfully conducting the detection. We observe that it is attributed to the secret key mostly used in the watermark detection -- it cannot be public, or the adversary may launch removal attacks provided the key; nor can it be private, or the watermarking detection is opaque to the public. To resolve the dilemma, we propose PVMark, a plugin based on zero-knowledge proof (ZKP), enabling the watermark detection process to be publicly verifiable by third parties without disclosing any secret key. PVMark hinges upon the proof of `correct execution' of watermark detection on which a set of ZKP constraints are built, including mapping, random number generation, comparison, and summation. We implement multiple variants of PVMark in Python, Rust and Circom, covering combinations of three watermarking schemes, three hash functions, and four ZKP protocols, to show our approach effectively works under a variety of circumstances. By experimental results, PVMark efficiently enables public verifiability on the state-of-the-art LLM watermarking schemes yet without compromising the watermarking performance, promising to be deployed in practice.

comment: This work has been submitted to the IEEE for possible publication

☆ Likely Interpolants of Generative Models

Frederik Möbius Rygaard, Shen Zhu, Yinzhu Jin, Søren Hauberg, Tom Fletcher

Interpolation in generative models allows for controlled generation, model inspection, and more. Unfortunately, most generative models lack a principal notion of interpolants without restrictive assumptions on either the model or data dimension. In this paper, we develop a general interpolation scheme that targets likely transition paths compatible with different metrics and probability distributions. We consider interpolants analogous to a geodesic constrained to a suitable data distribution and derive a novel algorithm for computing these curves, which requires no additional training. Theoretically, we show that our method locally can be considered as a geodesic under a suitable Riemannian metric. We quantitatively show that our interpolation scheme traverses higher density regions than baselines across a range of models and datasets.

☆ Angular Steering: Behavior Control via Rotation in Activation Space NeurIPS 2025

Hieu M. Vu, Tan M. Nguyen

Controlling specific behaviors in large language models while preserving their general capabilities is a central challenge for safe and reliable artificial intelligence deployment. Current steering methods, such as vector addition and directional ablation, are constrained within a two-dimensional subspace defined by the activation and feature direction, making them sensitive to chosen parameters and potentially affecting unrelated features due to unintended interactions in activation space. We introduce Angular Steering, a novel and flexible method for behavior modulation that operates by rotating activations within a fixed two-dimensional subspace. By formulating steering as a geometric rotation toward or away from a target behavior direction, Angular Steering provides continuous, fine-grained control over behaviors such as refusal and compliance. We demonstrate this method using refusal steering emotion steering as use cases. Additionally, we propose Adaptive Angular Steering, a selective variant that rotates only activations aligned with the target feature, further enhancing stability and coherence. Angular Steering generalizes existing addition and orthogonalization techniques under a unified geometric rotation framework, simplifying parameter selection and maintaining model stability across a broader range of adjustments. Experiments across multiple model families and sizes show that Angular Steering achieves robust behavioral control while maintaining general language modeling performance, underscoring its flexibility, generalization, and robustness compared to prior approaches. Code and artifacts are available at https://github.com/lone17/angular-steering/.

comment: NeurIPS 2025 (Spotlight)

☆ MPRU: Modular Projection-Redistribution Unlearning as Output Filter for Classification Pipelines

Minyi Peng, Darian Gunamardi, Ivan Tjuawinata, Kwok-Yan Lam

As a new and promising approach, existing machine unlearning (MU) works typically emphasize theoretical formulations or optimization objectives to achieve knowledge removal. However, when deployed in real-world scenarios, such solutions typically face scalability issues and have to address practical requirements such as full access to original datasets and model. In contrast to the existing approaches, we regard classification training as a sequential process where classes are learned sequentially, which we call \emph{inductive approach}. Unlearning can then be done by reversing the last training sequence. This is implemented by appending a projection-redistribution layer in the end of the model. Such an approach does not require full access to the original dataset or the model, addressing the challenges of existing methods. This enables modular and model-agnostic deployment as an output filter into existing classification pipelines with minimal alterations. We conducted multiple experiments across multiple datasets including image (CIFAR-10/100 using CNN-based model) and tabular datasets (Covertype using tree-based model). Experiment results show consistently similar output to a fully retrained model with a high computational cost reduction. This demonstrates the applicability, scalability, and system compatibility of our solution while maintaining the performance of the output in a more practical setting.

comment: 10 pages, 6 figures

☆ Test-Time Alignment of LLMs via Sampling-Based Optimal Control in pre-logit space

Sekitoshi Kanai, Tsukasa Yoshida, Hiroshi Takahashi, Haru Kuroki, Kazumune Hashimoto

Test-time alignment of large language models (LLMs) attracts attention because fine-tuning LLMs requires high computational costs. In this paper, we propose a new test-time alignment method called adaptive importance sampling on pre-logits (AISP) on the basis of the sampling-based model predictive control with the stochastic control input. AISP applies the Gaussian perturbation into pre-logits, which are outputs of the penultimate layer, so as to maximize expected rewards with respect to the mean of the perturbation. We demonstrate that the optimal mean is obtained by importance sampling with sampled rewards. AISP outperforms best-of-n sampling in terms of rewards over the number of used samples and achieves higher rewards than other reward-based test-time alignment methods.

comment: 21 pages, 8 figures

☆ Predicting All-Cause Hospital Readmissions from Medical Claims Data of Hospitalised Patients

Avinash Kadimisetty, Arun Rajagopalan, Vijendra SK

Reducing preventable hospital readmissions is a national priority for payers, providers, and policymakers seeking to improve health care and lower costs. The rate of readmission is being used as a benchmark to determine the quality of healthcare provided by the hospitals. In thisproject, we have used machine learning techniques like Logistic Regression, Random Forest and Support Vector Machines to analyze the health claims data and identify demographic and medical factors that play a crucial role in predicting all-cause readmissions. As the health claims data is high dimensional, we have used Principal Component Analysis as a dimension reduction technique and used the results for building regression models. We compared and evaluated these models based on the Area Under Curve (AUC) metric. Random Forest model gave the highest performance followed by Logistic Regression and Support Vector Machine models. These models can be used to identify the crucial factors causing readmissions and help identify patients to focus on to reduce the chances of readmission, ultimately bringing down the cost and increasing the quality of healthcare provided to the patients.

comment: NCMLAI 2018

☆ Accumulative SGD Influence Estimation for Data Attribution

Yunxiao Shi, Shuo Yang, Yixin Su, Rui Zhang, Min Xu

Modern data-centric AI needs precise per-sample influence. Standard SGD-IE approximates leave-one-out effects by summing per-epoch surrogates and ignores cross-epoch compounding, which misranks critical examples. We propose ACC-SGD-IE, a trajectory-aware estimator that propagates the leave-one-out perturbation across training and updates an accumulative influence state at each step. In smooth strongly convex settings it achieves geometric error contraction and, in smooth non-convex regimes, it tightens error bounds; larger mini-batches further reduce constants. Empirically, on Adult, 20 Newsgroups, and MNIST under clean and corrupted data and both convex and non-convex training, ACC-SGD-IE yields more accurate influence estimates, especially over long epochs. For downstream data cleansing it more reliably flags noisy samples, producing models trained on ACC-SGD-IE cleaned data that outperform those cleaned with SGD-IE.

☆ A Game-Theoretic Spatio-Temporal Reinforcement Learning Framework for Collaborative Public Resource Allocation

Songxin Lei, Qiongyan Wang, Yanchen Zhu, Hanyu Yao, Sijie Ruan, Weilin Ruan, Yuyu Luo, Huaming Wu, Yuxuan Liang

Public resource allocation involves the efficient distribution of resources, including urban infrastructure, energy, and transportation, to effectively meet societal demands. However, existing methods focus on optimizing the movement of individual resources independently, without considering their capacity constraints. To address this limitation, we propose a novel and more practical problem: Collaborative Public Resource Allocation (CPRA), which explicitly incorporates capacity constraints and spatio-temporal dynamics in real-world scenarios. We propose a new framework called Game-Theoretic Spatio-Temporal Reinforcement Learning (GSTRL) for solving CPRA. Our contributions are twofold: 1) We formulate the CPRA problem as a potential game and demonstrate that there is no gap between the potential function and the optimal target, laying a solid theoretical foundation for approximating the Nash equilibrium of this NP-hard problem; and 2) Our designed GSTRL framework effectively captures the spatio-temporal dynamics of the overall system. We evaluate GSTRL on two real-world datasets, where experiments show its superior performance. Our source codes are available in the supplementary materials.

☆ Segmentation over Complexity: Evaluating Ensemble and Hybrid Approaches for Anomaly Detection in Industrial Time Series

Emilio Mastriani, Alessandro Costa, Federico Incardona, Kevin Munari, Sebastiano Spinello

In this study, we investigate the effectiveness of advanced feature engineering and hybrid model architectures for anomaly detection in a multivariate industrial time series, focusing on a steam turbine system. We evaluate the impact of change point-derived statistical features, clustering-based substructure representations, and hybrid learning strategies on detection performance. Despite their theoretical appeal, these complex approaches consistently underperformed compared to a simple Random Forest + XGBoost ensemble trained on segmented data. The ensemble achieved an AUC-ROC of 0.976, F1-score of 0.41, and 100% early detection within the defined time window. Our findings highlight that, in scenarios with highly imbalanced and temporally uncertain data, model simplicity combined with optimized segmentation can outperform more sophisticated architectures, offering greater robustness, interpretability, and operational utility.

comment: This paper is currently under review for presentation at the IEEE SAMI 2026 Conference

☆ Bridging the Gap Between Molecule and Textual Descriptions via Substructure-aware Alignment EMNLP 2025

Hyuntae Park, Yeachan Kim, SangKeun Lee

Molecule and text representation learning has gained increasing interest due to its potential for enhancing the understanding of chemical information. However, existing models often struggle to capture subtle differences between molecules and their descriptions, as they lack the ability to learn fine-grained alignments between molecular substructures and chemical phrases. To address this limitation, we introduce MolBridge, a novel molecule-text learning framework based on substructure-aware alignments. Specifically, we augment the original molecule-description pairs with additional alignment signals derived from molecular substructures and chemical phrases. To effectively learn from these enriched alignments, MolBridge employs substructure-aware contrastive learning, coupled with a self-refinement mechanism that filters out noisy alignment signals. Experimental results show that MolBridge effectively captures fine-grained correspondences and outperforms state-of-the-art baselines on a wide range of molecular benchmarks, highlighting the significance of substructure-aware alignment in molecule-text learning.

comment: EMNLP 2025 (main)

☆ STAR: A Privacy-Preserving, Energy-Efficient Edge AI Framework for Human Activity Recognition via Wi-Fi CSI in Mobile and Pervasive Computing Environments

Kexing Liu

Human Activity Recognition (HAR) via Wi-Fi Channel State Information (CSI) presents a privacy-preserving, contactless sensing approach suitable for smart homes, healthcare monitoring, and mobile IoT systems. However, existing methods often encounter computational inefficiency, high latency, and limited feasibility within resource-constrained, embedded mobile edge environments. This paper proposes STAR (Sensing Technology for Activity Recognition), an edge-AI-optimized framework that integrates a lightweight neural architecture, adaptive signal processing, and hardware-aware co-optimization to enable real-time, energy-efficient HAR on low-power embedded devices. STAR incorporates a streamlined Gated Recurrent Unit (GRU)-based recurrent neural network, reducing model parameters by 33% compared to conventional LSTM models while maintaining effective temporal modeling capability. A multi-stage pre-processing pipeline combining median filtering, 8th-order Butterworth low-pass filtering, and Empirical Mode Decomposition (EMD) is employed to denoise CSI amplitude data and extract spatial-temporal features. For on-device deployment, STAR is implemented on a Rockchip RV1126 processor equipped with an embedded Neural Processing Unit (NPU), interfaced with an ESP32-S3-based CSI acquisition module. Experimental results demonstrate a mean recognition accuracy of 93.52% across seven activity classes and 99.11% for human presence detection, utilizing a compact 97.6k-parameter model. INT8 quantized inference achieves a processing speed of 33 MHz with just 8% CPU utilization, delivering sixfold speed improvements over CPU-based execution. With sub-second response latency and low power consumption, the system ensures real-time, privacy-preserving HAR, offering a practical, scalable solution for mobile and pervasive computing environments.

☆ maxVSTAR: Maximally Adaptive Vision-Guided CSI Sensing with Closed-Loop Edge Model Adaptation for Robust Human Activity Recognition

Kexing Liu

WiFi Channel State Information (CSI)-based human activity recognition (HAR) provides a privacy-preserving, device-free sensing solution for smart environments. However, its deployment on edge devices is severely constrained by domain shift, where recognition performance deteriorates under varying environmental and hardware conditions. This study presents maxVSTAR (maximally adaptive Vision-guided Sensing Technology for Activity Recognition), a closed-loop, vision-guided model adaptation framework that autonomously mitigates domain shift for edge-deployed CSI sensing systems. The proposed system integrates a cross-modal teacher-student architecture, where a high-accuracy YOLO-based vision model serves as a dynamic supervisory signal, delivering real-time activity labels for the CSI data stream. These labels enable autonomous, online fine-tuning of a lightweight CSI-based HAR model, termed Sensing Technology for Activity Recognition (STAR), directly at the edge. This closed-loop retraining mechanism allows STAR to continuously adapt to environmental changes without manual intervention. Extensive experiments demonstrate the effectiveness of maxVSTAR. When deployed on uncalibrated hardware, the baseline STAR model's recognition accuracy declined from 93.52% to 49.14%. Following a single vision-guided adaptation cycle, maxVSTAR restored the accuracy to 81.51%. These results confirm the system's capacity for dynamic, self-supervised model adaptation in privacy-conscious IoT environments, establishing a scalable and practical paradigm for long-term autonomous HAR using CSI sensing at the network edge.

☆ Beyond Synthetic Benchmarks: Evaluating LLM Performance on Real-World Class-Level Code Generation

Musfiqur Rahman, SayedHassan Khatoonabadi, Emad Shihab

Large language models (LLMs) have advanced code generation at the function level, yet their ability to produce correct class-level implementations in authentic software projects remains poorly understood. This work introduces a novel benchmark derived from open-source repositories, comprising real-world classes divided into seen and unseen partitions to evaluate generalization under practical conditions. The evaluation examines multiple LLMs under varied input specifications, retrieval-augmented configurations, and documentation completeness levels. Results reveal a stark performance disparity: LLMs achieve 84% to 89% correctness on established synthetic benchmarks but only 25% to 34% on real-world class tasks, with negligible differences between familiar and novel codebases. Comprehensive docstrings yield modest gains of 1% to 3% in functional accuracy, though statistical significance is rare. Retrieval-augmented generation proves most effective with partial documentation, improving correctness by 4% to 7% by supplying concrete implementation patterns absent from specifications. Error profiling identifies AttributeError, TypeError, and AssertionError as dominant failure modes (84% of cases), with synthetic tests overemphasizing assertion issues and real-world scenarios highlighting type and attribute mismatches. Retrieval augmentation reduces logical flaws but can introduce dependency conflicts. The benchmark and analysis expose critical limitations in current LLM capabilities for class-level engineering, offering actionable insights for enhancing context modelling, documentation strategies, and retrieval integration in production code assistance tools.

comment: Pre-print prepared for journal submission

☆ Uncertainty-Aware Diagnostics for Physics-Informed Machine Learning

Mara Daniels, Liam Hodgkinson, Michael Mahoney

Physics-informed machine learning (PIML) integrates prior physical information, often in the form of differential equation constraints, into the process of fitting machine learning models to physical data. Popular PIML approaches, including neural operators, physics-informed neural networks, neural ordinary differential equations, and neural discrete equilibria, are typically fit to objectives that simultaneously include both data and physical constraints. However, the multi-objective nature of this approach creates ambiguity in the measurement of model quality. This is related to a poor understanding of epistemic uncertainty, and it can lead to surprising failure modes, even when existing statistical metrics suggest strong fits. Working within a Gaussian process regression framework, we introduce the Physics-Informed Log Evidence (PILE) score. Bypassing the ambiguities of test losses, the PILE score is a single, uncertainty-aware metric that provides a selection principle for hyperparameters of a PIML model. We show that PILE minimization yields excellent choices for a wide variety of model parameters, including kernel bandwidth, least squares regularization weights, and even kernel function selection. We also show that, even prior to data acquisition, a special 'data-free' case of the PILE score identifies a priori kernel choices that are 'well-adapted' to a given PDE. Beyond the kernel setting, we anticipate that the PILE score can be extended to PIML at large, and we outline approaches to do so.

☆ Do Not Step Into the Same River Twice: Learning to Reason from Trial and Error

Chenming Tang, Hsiu-Yuan Huang, Weijie Liu, Saiyong Yang, Yunfang Wu

Reinforcement learning with verifiable rewards (RLVR) has significantly boosted the reasoning capability of large language models (LLMs) recently. However, existing RLVR approaches merely train LLMs based on their own generated responses and are constrained by the initial capability of LLMs, thus prone to exploration stagnation, in which LLMs fail to solve more training problems and cannot further learn from the training data. Some work tries to address this by leveraging off-policy solutions to training problems but requires external guidance from experts which suffers from limited availability. In this work, we propose LTE (Learning to reason from Trial and Error), an approach hinting LLMs with their previously self-generated incorrect answers and problem of overlong responses, which does not require any external expert guidance. Experiments validate the effectiveness of LTE, which outperforms the normal group relative policy optimization (GRPO) by 6.38 in Pass@1 and 9.00 in Pass@k on average across six mathematics benchmarks for Qwen3-4B-Base. Further analysis confirms that LTE successfully mitigates the problem of exploration stagnation and enhances both exploitation and exploration during training.

comment: Work in progress

☆ SAFE: A Novel Approach to AI Weather Evaluation through Stratified Assessments of Forecasts over Earth

Nick Masi, Randall Balestriero

The dominant paradigm in machine learning is to assess model performance based on average loss across all samples in some test set. This amounts to averaging performance geospatially across the Earth in weather and climate settings, failing to account for the non-uniform distribution of human development and geography. We introduce Stratified Assessments of Forecasts over Earth (SAFE), a package for elucidating the stratified performance of a set of predictions made over Earth. SAFE integrates various data domains to stratify by different attributes associated with geospatial gridpoints: territory (usually country), global subregion, income, and landcover (land or water). This allows us to examine the performance of models for each individual stratum of the different attributes (e.g., the accuracy in every individual country). To demonstrate its importance, we utilize SAFE to benchmark a zoo of state-of-the-art AI-based weather prediction models, finding that they all exhibit disparities in forecasting skill across every attribute. We use this to seed a benchmark of model forecast fairness through stratification at different lead times for various climatic variables. By moving beyond globally-averaged metrics, we for the first time ask: where do models perform best or worst, and which models are most fair? To support further work in this direction, the SAFE package is open source and available at https://github.com/N-Masi/safe

☆ Robust Super-Capacity SRS Channel Inpainting via Diffusion Models

Usman Akram, Fan Zhang, Yang Li, Haris Vikalo

Accurate channel state information (CSI) is essential for reliable multiuser MIMO operation. In 5G NR, reciprocity-based beamforming via uplink Sounding Reference Signals (SRS) face resource and coverage constraints, motivating sparse non-uniform SRS allocation. Prior masked-autoencoder (MAE) approaches improve coverage but overfit to training masks and degrade under unseen distortions (e.g., additional masking, interference, clipping, non-Gaussian noise). We propose a diffusion-based channel inpainting framework that integrates system-model knowledge at inference via a likelihood-gradient term, enabling a single trained model to adapt across mismatched conditions. On standardized CDL channels, the score-based diffusion variant consistently outperforms a UNet score-model baseline and the one-step MAE under distribution shift, with improvements up to 14 dB NMSE in challenging settings (e.g., Laplace noise, user interference), while retaining competitive accuracy under matched conditions. These results demonstrate that diffusion-guided inpainting is a robust and generalizable approach for super-capacity SRS design in 5G NR systems.

☆ ALMGuard: Safety Shortcuts and Where to Find Them as Guardrails for Audio-Language Models NeurIPS 2025

Weifei Jin, Yuxin Cao, Junjie Su, Minhui Xue, Jie Hao, Ke Xu, Jin Song Dong, Derui Wang

Recent advances in Audio-Language Models (ALMs) have significantly improved multimodal understanding capabilities. However, the introduction of the audio modality also brings new and unique vulnerability vectors. Previous studies have proposed jailbreak attacks that specifically target ALMs, revealing that defenses directly transferred from traditional audio adversarial attacks or text-based Large Language Model (LLM) jailbreaks are largely ineffective against these ALM-specific threats. To address this issue, we propose ALMGuard, the first defense framework tailored to ALMs. Based on the assumption that safety-aligned shortcuts naturally exist in ALMs, we design a method to identify universal Shortcut Activation Perturbations (SAPs) that serve as triggers that activate the safety shortcuts to safeguard ALMs at inference time. To better sift out effective triggers while preserving the model's utility on benign tasks, we further propose Mel-Gradient Sparse Mask (M-GSM), which restricts perturbations to Mel-frequency bins that are sensitive to jailbreaks but insensitive to speech understanding. Both theoretical analyses and empirical results demonstrate the robustness of our method against both seen and unseen attacks. Overall, \MethodName reduces the average success rate of advanced ALM-specific jailbreak attacks to 4.6% across four models, while maintaining comparable utility on benign benchmarks, establishing it as the new state of the art. Our code and data are available at https://github.com/WeifeiJin/ALMGuard.

comment: Accepted to NeurIPS 2025

☆ Lean4Physics: Comprehensive Reasoning Framework for College-level Physics in Lean4

Yuxin Li, Minghao Liu, Ruida Wang, Wenzhao Ji, Zhitao He, Rui Pan, Junming Huang, Tong Zhang, Yi R. Fung

We present **Lean4PHYS**, a comprehensive reasoning framework for college-level physics problems in Lean4. **Lean4PHYS** includes *LeanPhysBench*, a college-level benchmark for formal physics reasoning in Lean4, which contains 200 hand-crafted and peer-reviewed statements derived from university textbooks and physics competition problems. To establish a solid foundation for formal reasoning in physics, we also introduce *PhysLib*, a community-driven repository containing fundamental unit systems and theorems essential for formal physics reasoning. Based on the benchmark and Lean4 repository we composed in **Lean4PHYS**, we report baseline results using major expert Math Lean4 provers and state-of-the-art closed-source models, with the best performance of DeepSeek-Prover-V2-7B achieving only 16% and Claude-Sonnet-4 achieving 35%. We also conduct a detailed analysis showing that our *PhysLib* can achieve an average improvement of 11.75% in model performance. This demonstrates the challenging nature of our *LeanPhysBench* and the effectiveness of *PhysLib*. To the best of our knowledge, this is the first study to provide a physics benchmark in Lean4.

☆ Network-Constrained Policy Optimization for Adaptive Multi-agent Vehicle Routing

Fazel Arasteh, Arian Haghparast, Manos Papagelis

Traffic congestion in urban road networks leads to longer trip times and higher emissions, especially during peak periods. While the Shortest Path First (SPF) algorithm is optimal for a single vehicle in a static network, it performs poorly in dynamic, multi-vehicle settings, often worsening congestion by routing all vehicles along identical paths. We address dynamic vehicle routing through a multi-agent reinforcement learning (MARL) framework for coordinated, network-aware fleet navigation. We first propose Adaptive Navigation (AN), a decentralized MARL model where each intersection agent provides routing guidance based on (i) local traffic and (ii) neighborhood state modeled using Graph Attention Networks (GAT). To improve scalability in large networks, we further propose Hierarchical Hub-based Adaptive Navigation (HHAN), an extension of AN that assigns agents only to key intersections (hubs). Vehicles are routed hub-to-hub under agent control, while SPF handles micro-routing within each hub region. For hub coordination, HHAN adopts centralized training with decentralized execution (CTDE) under the Attentive Q-Mixing (A-QMIX) framework, which aggregates asynchronous vehicle decisions via attention. Hub agents use flow-aware state features that combine local congestion and predictive dynamics for proactive routing. Experiments on synthetic grids and real urban maps (Toronto, Manhattan) show that AN reduces average travel time versus SPF and learning baselines, maintaining 100% routing success. HHAN scales to networks with hundreds of intersections, achieving up to 15.9% improvement under heavy traffic. These findings highlight the potential of network-constrained MARL for scalable, coordinated, and congestion-aware routing in intelligent transportation systems.

comment: 29 pages, 12 figures. Fazel Arasteh and Arian Haghparast contributed equally to this research. Submitted to ACM Transactions on Spatial Algorithms and Systems (TSAS). The code for this work is publicly available at https://github.com/Arianhgh/HHAN

☆ LLMBisect: Breaking Barriers in Bug Bisection with A Comparative Analysis Pipeline

Zheng Zhang, Haonan Li, Xingyu Li, Hang Zhang, Zhiyun Qian

Bug bisection has been an important security task that aims to understand the range of software versions impacted by a bug, i.e., identifying the commit that introduced the bug. However, traditional patch-based bisection methods are faced with several significant barriers: For example, they assume that the bug-inducing commit (BIC) and the patch commit modify the same functions, which is not always true. They often rely solely on code changes, while the commit message frequently contains a wealth of vulnerability-related information. They are also based on simple heuristics (e.g., assuming the BIC initializes lines deleted in the patch) and lack any logical analysis of the vulnerability. In this paper, we make the observation that Large Language Models (LLMs) are well-positioned to break the barriers of existing solutions, e.g., comprehend both textual data and code in patches and commits. Unlike previous BIC identification approaches, which yield poor results, we propose a comprehensive multi-stage pipeline that leverages LLMs to: (1) fully utilize patch information, (2) compare multiple candidate commits in context, and (3) progressively narrow down the candidates through a series of down-selection steps. In our evaluation, we demonstrate that our approach achieves significantly better accuracy than the state-of-the-art solution by more than 38\%. Our results further confirm that the comprehensive multi-stage pipeline is essential, as it improves accuracy by 60\% over a baseline LLM-based bisection method.

☆ Nirvana: A Specialized Generalist Model With Task-Aware Memory Mechanism

Yuhua Jiang, Shuang Cheng, Yihao Liu, Ermo Hua, Che Jiang, Weigao Sun, Yu Cheng, Feifei Gao, Biqing Qi, Bowen Zhou

Specialized Generalist Models (SGMs) aim to preserve broad capabilities while achieving expert-level performance in target domains. However, traditional LLM structures including Transformer, Linear Attention, and hybrid models do not employ specialized memory mechanism guided by task information. In this paper, we present Nirvana, an SGM with specialized memory mechanism, linear time complexity, and test-time task information extraction. Besides, we propose the Task-Aware Memory Trigger ($\textit{Trigger}$) that flexibly adjusts memory mechanism based on the current task's requirements. In Trigger, each incoming sample is treated as a self-supervised fine-tuning task, enabling Nirvana to adapt its task-related parameters on the fly to domain shifts. We also design the Specialized Memory Updater ($\textit{Updater}$) that dynamically memorizes the context guided by Trigger. We conduct experiments on both general language tasks and specialized medical tasks. On a variety of natural language modeling benchmarks, Nirvana achieves competitive or superior results compared to the existing LLM structures. To prove the effectiveness of Trigger on specialized tasks, we test Nirvana's performance on a challenging medical task, i.e., Magnetic Resonance Imaging (MRI). We post-train frozen Nirvana backbone with lightweight codecs on paired electromagnetic signals and MRI images. Despite the frozen Nirvana backbone, Trigger guides the model to adapt to the MRI domain with the change of task-related parameters. Nirvana achieves higher-quality MRI reconstruction compared to conventional MRI models as well as the models with traditional LLMs' backbone, and can also generate accurate preliminary clinical reports accordingly.

♻ ☆ UniSite: The First Cross-Structure Dataset and Learning Framework for End-to-End Ligand Binding Site Detection

Jigang Fan, Quanlin Wu, Shengjie Luo, Liwei Wang

The detection of ligand binding sites for proteins is a fundamental step in Structure-Based Drug Design. Despite notable advances in recent years, existing methods, datasets, and evaluation metrics are confronted with several key challenges: (1) current datasets and methods are centered on individual protein-ligand complexes and neglect that diverse binding sites may exist across multiple complexes of the same protein, introducing significant statistical bias; (2) ligand binding site detection is typically modeled as a discontinuous workflow, employing binary segmentation and subsequent clustering algorithms; (3) traditional evaluation metrics do not adequately reflect the actual performance of different binding site prediction methods. To address these issues, we first introduce UniSite-DS, the first UniProt (Unique Protein)-centric ligand binding site dataset, which contains 4.81 times more multi-site data and 2.08 times more overall data compared to the previously most widely used datasets. We then propose UniSite, the first end-to-end ligand binding site detection framework supervised by set prediction loss with bijective matching. In addition, we introduce Average Precision based on Intersection over Union (IoU) as a more accurate evaluation metric for ligand binding site prediction. Extensive experiments on UniSite-DS and several representative benchmark datasets demonstrate that IoU-based Average Precision provides a more accurate reflection of prediction quality, and that UniSite outperforms current state-of-the-art methods in ligand binding site detection. The dataset and codes will be made publicly available at https://github.com/quanlin-wu/unisite.

♻ ☆ S-CFE: Simple Counterfactual Explanations

Shpresim Sadiku, Moritz Wagner, Sai Ganesh Nagarajan, Sebastian Pokutta

We study the problem of finding optimal sparse, manifold-aligned counterfactual explanations for classifiers. Canonically, this can be formulated as an optimization problem with multiple non-convex components, including classifier loss functions and manifold alignment (or \emph{plausibility}) metrics. The added complexity of enforcing \emph{sparsity}, or shorter explanations, complicates the problem further. Existing methods often focus on specific models and plausibility measures, relying on convex $\ell_1$ regularizers to enforce sparsity. In this paper, we tackle the canonical formulation using the accelerated proximal gradient (APG) method, a simple yet efficient first-order procedure capable of handling smooth non-convex objectives and non-smooth $\ell_p$ (where $0 \leq p < 1$) regularizers. This enables our approach to seamlessly incorporate various classifiers and plausibility measures while producing sparser solutions. Our algorithm only requires differentiable data-manifold regularizers and supports box constraints for bounded feature ranges, ensuring the generated counterfactuals remain \emph{actionable}. Finally, experiments on real-world datasets demonstrate that our approach effectively produces sparse, manifold-aligned counterfactual explanations while maintaining proximity to the factual data and computational efficiency.

♻ ☆ Direct Debiased Machine Learning via Bregman Divergence Minimization

Masahiro Kato

We develop a direct debiased machine learning framework comprising Neyman targeted estimation and generalized Riesz regression. Our framework unifies Riesz regression for automatic debiased machine learning, covariate balancing, targeted maximum likelihood estimation (TMLE), and density-ratio estimation. In many problems involving causal effects or structural models, the parameters of interest depend on regression functions. Plugging regression functions estimated by machine learning methods into the identifying equations can yield poor performance because of first-stage bias. To reduce such bias, debiased machine learning employs Neyman orthogonal estimating equations. Debiased machine learning typically requires estimation of the Riesz representer and the regression function. For this problem, we develop a direct debiased machine learning framework with an end-to-end algorithm. We formulate estimation of the nuisance parameters, the regression function and the Riesz representer, as minimizing the discrepancy between Neyman orthogonal scores computed with known and unknown nuisance parameters, which we refer to as Neyman targeted estimation. Neyman targeted estimation includes Riesz representer estimation, and we measure discrepancies using the Bregman divergence. The Bregman divergence encompasses various loss functions as special cases, where the squared loss yields Riesz regression and the Kullback-Leibler divergence yields entropy balancing. We refer to this Riesz representer estimation as generalized Riesz regression. Neyman targeted estimation also yields TMLE as a special case for regression function estimation. Furthermore, for specific pairs of models and Riesz representer estimation methods, we can automatically obtain the covariate balancing property without explicitly solving the covariate balancing objective.

♻ ☆ Adversarial generalization of unfolding (model-based) networks NeurIPS2025

Vicky Kouni

Unfolding networks are interpretable networks emerging from iterative algorithms, incorporate prior knowledge of data structure, and are designed to solve inverse problems like compressed sensing, which deals with recovering data from noisy, missing observations. Compressed sensing finds applications in critical domains, from medical imaging to cryptography, where adversarial robustness is crucial to prevent catastrophic failures. However, a solid theoretical understanding of the performance of unfolding networks in the presence of adversarial attacks is still in its infancy. In this paper, we study the adversarial generalization of unfolding networks when perturbed with $l_2$-norm constrained attacks, generated by the fast gradient sign method. Particularly, we choose a family of state-of-the-art overaparameterized unfolding networks and deploy a new framework to estimate their adversarial Rademacher complexity. Given this estimate, we provide adversarial generalization error bounds for the networks under study, which are tight with respect to the attack level. To our knowledge, this is the first theoretical analysis on the adversarial generalization of unfolding networks. We further present a series of experiments on real-world data, with results corroborating our derived theory, consistently for all data. Finally, we observe that the family's overparameterization can be exploited to promote adversarial robustness, shedding light on how to efficiently robustify neural networks.

comment: Accepted at NeurIPS2025

♻ ☆ Advancing Local Clustering on Graphs via Compressive Sensing: Semi-supervised and Unsupervised Methods

Zhaiming Shen, Sung Ha Kang

Local clustering aims to identify specific substructures within a large graph without any additional structural information of the graph. These substructures are typically small compared to the overall graph, enabling the problem to be approached by finding a sparse solution to a linear system associated with the graph Laplacian. In this work, we first propose a method for identifying specific local clusters when very few labeled data are given, which we term semi-supervised local clustering. We then extend this approach to the unsupervised setting when no prior information on labels is available. The proposed methods involve randomly sampling the graph, applying diffusion through local cluster extraction, then examining the overlap among the results to find each cluster. We establish the co-membership conditions for any pair of nodes, and rigorously prove the correctness of our methods. Additionally, we conduct extensive experiments to demonstrate that the proposed methods achieve state of the art results in the low-label rates regime.

♻ ☆ Partially-Supervised Neural Network Model For Quadratic Multiparametric Programming

Fuat Can Beylunioglu, Mehrdad Pirnia, P. Robert Duimering

Neural Networks (NN) with ReLU activation functions are used to model multiparametric quadratic optimization problems (mp-QP) in diverse engineering applications. Researchers have suggested leveraging the piecewise affine property of deep NN models to solve mp-QP with linear constraints, which also exhibit piecewise affine behaviour. However, traditional deep NN applications to mp-QP fall short of providing optimal and feasible predictions, even when trained on large datasets. This study proposes a partially-supervised NN (PSNN) architecture that directly represents the mathematical structure of the global solution function. In contrast to generic NN training approaches, the proposed PSNN method derives a large proportion of model weights directly from the mathematical properties of the optimization problem, producing more accurate solutions despite significantly smaller training data sets. Many energy management problems are formulated as QP, so we apply the proposed approach to energy systems (specifically DC optimal power flow) to demonstrate proof of concept. Model performance in terms of solution accuracy and speed of predictions was compared against a commercial solver and a generic Deep NN model based on classical training. Results show KKT sufficient conditions for PSNN consistently outperform generic NN architectures with classical training using far less data, including when tested on extreme, out-of-training distribution test data. Given its speed advantages over traditional solvers, the PSNN model can quickly produce optimal and feasible solutions within a second for millions of input parameters sampled from a distribution of stochastic demands and renewable generator dispatches, which can be used for simulations and long term planning.

comment: 36 pages including references and appendix

♻ ☆ Integrating Protein Sequence and Expression Level to Analysis Molecular Characterization of Breast Cancer Subtypes

Hossein Sholehrasa, Majid Jaberi-Douraki

Breast cancer's complexity and variability pose significant challenges in understanding its progression and guiding effective treatment. This study aims to integrate protein sequence data with expression levels to improve the molecular characterization of breast cancer subtypes and predict clinical outcomes. Using ProtGPT2, a language model specifically designed for protein sequences, we generated embeddings that capture the functional and structural properties of proteins. These embeddings were integrated with protein expression levels to form enriched biological representations, which were analyzed using machine learning methods, such as ensemble K-means for clustering and XGBoost for classification. Our approach enabled the successful clustering of patients into biologically distinct groups and accurately predicted clinical outcomes such as survival and biomarker status, achieving high performance metrics, notably an F1 score of 0.88 for survival and 0.87 for biomarker status prediction. Feature importance analysis identified KMT2C, CLASP2, and MYO1B as key proteins involved in hormone signaling, cytoskeletal remodeling, and therapy resistance in hormone receptor-positive and triple-negative breast cancer, with potential influence on breast cancer subtype behavior and progression. Furthermore, protein-protein interaction networks and correlation analyses revealed functional interdependencies among proteins that may influence the behavior and progression of breast cancer subtypes. These findings suggest that integrating protein sequence and expression data provides valuable insights into tumor biology and has significant potential to enhance personalized treatment strategies in breast cancer care.

♻ ☆ When Kernels Multiply, Clusters Unify: Fusing Embeddings with the Kronecker Product

Youqi Wu, Jingwei Zhang, Farzan Farnia

State-of-the-art embeddings often capture distinct yet complementary discriminative features: For instance, one image embedding model may excel at distinguishing fine-grained textures, while another focuses on object-level structure. Motivated by this observation, we propose a principled approach to fuse such complementary representations through kernel multiplication. Multiplying the kernel similarity functions of two embeddings allows their discriminative structures to interact, producing a fused representation whose kernel encodes the union of the clusters identified by each parent embedding. This formulation also provides a natural way to construct joint kernels for paired multi-modal data (e.g., image-text tuples), where the product of modality-specific kernels inherits structure from both domains. We highlight that this kernel product is mathematically realized via the Kronecker product of the embedding feature maps, yielding our proposed KrossFuse framework for embedding fusion. To address the computational cost of the resulting high-dimensional Kronecker space, we further develop RP-KrossFuse, a scalable variant that leverages random projections for efficient approximation. As a key application, we use this framework to bridge the performance gap between cross-modal embeddings (e.g., CLIP, BLIP) and unimodal experts (e.g., DINOv2, E5). Experiments show that RP-KrossFuse effectively integrates these models, enhancing modality-specific performance while preserving cross-modal alignment. The project code is available at https://github.com/yokiwuuu/KrossFuse.

♻ ☆ RLBFF: Binary Flexible Feedback to bridge between Human Feedback & Verifiable Rewards

Zhilin Wang, Jiaqi Zeng, Olivier Delalleau, Ellie Evans, Daniel Egert, Hoo-Chang Shin, Felipe Soares, Yi Dong, Oleksii Kuchaiev

Reinforcement Learning with Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) are the main RL paradigms used in LLM post-training, each offering distinct advantages. However, RLHF struggles with interpretability and reward hacking because it relies on human judgments that usually lack explicit criteria, whereas RLVR is limited in scope by its focus on correctness-based verifiers. We propose Reinforcement Learning with Binary Flexible Feedback (RLBFF), which combines the versatility of human-driven preferences with the precision of rule-based verification, enabling reward models to capture nuanced aspects of response quality beyond mere correctness. RLBFF extracts principles that can be answered in a binary fashion (e.g. accuracy of information: yes, or code readability: no) from natural language feedback. Such principles can then be used to ground Reward Model training as an entailment task (response satisfies or does not satisfy an arbitrary principle). We show that Reward Models trained in this manner can outperform Bradley-Terry models when matched for data and achieve top performance on RM-Bench (86.2%) and JudgeBench (81.4%, #1 on leaderboard as of September 24, 2025). Additionally, users can specify principles of interest at inference time to customize the focus of our reward models, in contrast to Bradley-Terry models. Finally, we present a fully open source recipe (including data) to align Qwen3-32B using RLBFF and our Reward Model, to match or exceed the performance of o3-mini and DeepSeek R1 on general alignment benchmarks of MT-Bench, WildBench, and Arena Hard v2 (at <5% of the inference cost). Models: https://huggingface.co/collections/nvidia/reward-models-10-2025

comment: Added link to access models: https://huggingface.co/collections/nvidia/reward-models-10-2025

♻ ☆ Curriculum Abductive Learning NeurIPS 2025

Wen-Chao Hu, Qi-Jie Li, Lin-Han Jia, Cunjing Ge, Yu-Feng Li, Yuan Jiang, Zhi-Hua Zhou

Abductive Learning (ABL) integrates machine learning with logical reasoning in a loop: a learning model predicts symbolic concept labels from raw inputs, which are revised through abduction using domain knowledge and then fed back for retraining. However, due to the nondeterminism of abduction, the training process often suffers from instability, especially when the knowledge base is large and complex, resulting in a prohibitively large abduction space. While prior works focus on improving candidate selection within this space, they typically treat the knowledge base as a static black box. In this work, we propose Curriculum Abductive Learning (C-ABL), a method that explicitly leverages the internal structure of the knowledge base to address the ABL training challenges. C-ABL partitions the knowledge base into a sequence of sub-bases, progressively introduced during training. This reduces the abduction space throughout training and enables the model to incorporate logic in a stepwise, smooth way. Experiments across multiple tasks show that C-ABL outperforms previous ABL implementations, significantly improves training stability, convergence speed, and final accuracy, especially under complex knowledge setting.

comment: Accepted by NeurIPS 2025, 22 pages, 6 figures

♻ ☆ Guided Model Merging for Hybrid Data Learning: Leveraging Centralized Data to Refine Decentralized Models WACV 2026

Junyi Zhu, Ruicong Yao, Taha Ceritli, Savas Ozkan, Matthew B. Blaschko, Eunchung Noh, Jeongwon Min, Cho Jung Min, Mete Ozay

Current network training paradigms primarily focus on either centralized or decentralized data regimes. However, in practice, data availability often exhibits a hybrid nature, where both regimes coexist. This hybrid setting presents new opportunities for model training, as the two regimes offer complementary trade-offs: decentralized data is abundant but subject to heterogeneity and communication constraints, while centralized data, though limited in volume and potentially unrepresentative, enables better curation and high-throughput access. Despite its potential, effectively combining these paradigms remains challenging, and few frameworks are tailored to hybrid data regimes. To address this, we propose a novel framework that constructs a model atlas from decentralized models and leverages centralized data to refine a global model within this structured space. The refined model is then used to reinitialize the decentralized models. Our method synergizes federated learning (to exploit decentralized data) and model merging (to utilize centralized data), enabling effective training under hybrid data availability. Theoretically, we show that our approach achieves faster convergence than methods relying solely on decentralized data, due to variance reduction in the merging process. Extensive experiments demonstrate that our framework consistently outperforms purely centralized, purely decentralized, and existing hybrid-adaptable methods. Notably, our method remains robust even when the centralized and decentralized data domains differ or when decentralized data contains noise, significantly broadening its applicability.

comment: Accepted at WACV 2026

♻ ☆ GSE: Group-wise Sparse and Explainable Adversarial Attacks

Shpresim Sadiku, Moritz Wagner, Sebastian Pokutta

Sparse adversarial attacks fool deep neural networks (DNNs) through minimal pixel perturbations, often regularized by the $\ell_0$ norm. Recent efforts have replaced this norm with a structural sparsity regularizer, such as the nuclear group norm, to craft group-wise sparse adversarial attacks. The resulting perturbations are thus explainable and hold significant practical relevance, shedding light on an even greater vulnerability of DNNs. However, crafting such attacks poses an optimization challenge, as it involves computing norms for groups of pixels within a non-convex objective. We address this by presenting a two-phase algorithm that generates group-wise sparse attacks within semantically meaningful areas of an image. Initially, we optimize a quasinorm adversarial loss using the $1/2-$quasinorm proximal operator tailored for non-convex programming. Subsequently, the algorithm transitions to a projected Nesterov's accelerated gradient descent with $2-$norm regularization applied to perturbation magnitudes. Rigorous evaluations on CIFAR-10 and ImageNet datasets demonstrate a remarkable increase in group-wise sparsity, e.g., $50.9\%$ on CIFAR-10 and $38.4\%$ on ImageNet (average case, targeted attack). This performance improvement is accompanied by significantly faster computation times, improved explainability, and a $100\%$ attack success rate.

♻ ☆ Understanding Generalization in Node and Link Prediction

Antonis Vasileiou, Timo Stoll, Christopher Morris

Using message-passing graph neural networks (MPNNs) for node and link prediction is crucial in various scientific and industrial domains, which has led to the development of diverse MPNN architectures. Besides working well in practical settings, their ability to generalize beyond the training set remains poorly understood. While some studies have explored MPNNs' generalization in graph-level prediction tasks, much less attention has been given to node- and link-level predictions. Existing works often rely on unrealistic i.i.d.\@ assumptions, overlooking possible correlations between nodes or links, and assuming fixed aggregation and impractical loss functions while neglecting the influence of graph structure. In this work, we introduce a unified framework to analyze the generalization properties of MPNNs in inductive and transductive node and link prediction settings, incorporating diverse architectural parameters and loss functions and quantifying the influence of graph structure. Additionally, our proposed generalization framework can be applied beyond graphs to any classification task under the inductive or transductive setting. Our empirical study supports our theoretical insights, deepening our understanding of MPNNs' generalization capabilities in these tasks.

comment: arXiv admin note: text overlap with arXiv:2412.07106

♻ ☆ RelP: Faithful and Efficient Circuit Discovery in Language Models via Relevance Patching

Farnoush Rezaei Jafari, Oliver Eberle, Ashkan Khakzar, Neel Nanda

Activation patching is a standard method in mechanistic interpretability for localizing the components of a model responsible for specific behaviors, but it is computationally expensive to apply at scale. Attribution patching offers a faster, gradient-based approximation, yet suffers from noise and reduced reliability in deep, highly non-linear networks. In this work, we introduce Relevance Patching (RelP), which replaces the local gradients in attribution patching with propagation coefficients derived from Layer-wise Relevance Propagation (LRP). LRP propagates the network's output backward through the layers, redistributing relevance to lower-level components according to local propagation rules that ensure properties such as relevance conservation or improved signal-to-noise ratio. Like attribution patching, RelP requires only two forward passes and one backward pass, maintaining computational efficiency while improving faithfulness. We validate RelP across a range of models and tasks, showing that it more accurately approximates activation patching than standard attribution patching, particularly when analyzing residual stream and MLP outputs in the Indirect Object Identification (IOI) task. For instance, for MLP outputs in GPT-2 Large, attribution patching achieves a Pearson correlation of 0.006, whereas RelP reaches 0.956, highlighting the improvement offered by RelP. Additionally, we compare the faithfulness of sparse feature circuits identified by RelP and Integrated Gradients (IG), showing that RelP achieves comparable faithfulness without the extra computational cost associated with IG.

♻ ☆ PepCompass: Navigating peptide embedding spaces using Riemannian Geometry

Marcin Możejko, Adam Bielecki, Jurand Prądzyński, Marcin Traskowski, Antoni Janowski, Hyun-Su Lee, Marcelo Der Torossian Torres, Michał Kmicikiewicz, Paulina Szymczak, Karol Jurasz, Michał Kucharczyk, Cesar de la Fuente-Nunez, Ewa Szczurek

Antimicrobial peptide discovery is challenged by the astronomical size of peptide space and the relative scarcity of active peptides. Generative models provide continuous latent "maps" of peptide space, but conventionally ignore decoder-induced geometry and rely on flat Euclidean metrics, rendering exploration and optimization distorted and inefficient. Prior manifold-based remedies assume fixed intrinsic dimensionality, which critically fails in practice for peptide data. Here, we introduce PepCompass, a geometry-aware framework for peptide exploration and optimization. At its core, we define a Union of $\kappa$-Stable Riemannian Manifolds $\mathbb{M}^{\kappa}$, a family of decoder-induced manifolds that captures local geometry while ensuring computational stability. We propose two local exploration methods: Second-Order Riemannian Brownian Efficient Sampling, which provides a convergent second-order approximation to Riemannian Brownian motion, and Mutation Enumeration in Tangent Space, which reinterprets tangent directions as discrete amino-acid substitutions. Combining these yields Local Enumeration Bayesian Optimization (LE-BO), an efficient algorithm for local activity optimization. Finally, we introduce Potential-minimizing Geodesic Search (PoGS), which interpolates between prototype embeddings along property-enriched geodesics, biasing discovery toward seeds, i.e. peptides with favorable activity. In-vitro validation confirms the effectiveness of PepCompass: PoGS yields four novel seeds, and subsequent optimization with LE-BO discovers 25 highly active peptides with broad-spectrum activity, including against resistant bacterial strains. These results demonstrate that geometry-informed exploration provides a powerful new paradigm for antimicrobial peptide design.

♻ ☆ C-LoRA: Contextual Low-Rank Adaptation for Uncertainty Estimation in Large Language Models NeurIPS 2025

Amir Hossein Rahmati, Sanket Jantre, Weifeng Zhang, Yucheng Wang, Byung-Jun Yoon, Nathan M. Urban, Xiaoning Qian

Low-Rank Adaptation (LoRA) offers a cost-effective solution for fine-tuning large language models (LLMs), but it often produces overconfident predictions in data-scarce few-shot settings. To address this issue, several classical statistical learning approaches have been repurposed for scalable uncertainty-aware LoRA fine-tuning. However, these approaches neglect how input characteristics affect the predictive uncertainty estimates. To address this limitation, we propose Contextual Low-Rank Adaptation (C-LoRA) as a novel uncertainty-aware and parameter efficient fine-tuning approach, by developing new lightweight LoRA modules contextualized to each input data sample to dynamically adapt uncertainty estimates. Incorporating data-driven contexts into the parameter posteriors, C-LoRA mitigates overfitting, achieves well-calibrated uncertainties, and yields robust predictions. Extensive experiments on LLaMA2-7B models demonstrate that C-LoRA consistently outperforms the state-of-the-art uncertainty-aware LoRA methods in both uncertainty quantification and model generalization. Ablation studies further confirm the critical role of our contextual modules in capturing sample-specific uncertainties. C-LoRA sets a new standard for robust, uncertainty-aware LLM fine-tuning in few-shot regimes. Although our experiments are limited to 7B models, our method is architecture-agnostic and, in principle, applies beyond this scale; studying its scaling to larger models remains an open problem. Our code is available at https://github.com/ahra99/c_lora.

comment: Conference on Neural Information Processing Systems (NeurIPS 2025)

♻ ☆ MaskCaptioner: Learning to Jointly Segment and Caption Object Trajectories in Videos

Gabriel Fiastre, Antoine Yang, Cordelia Schmid

Dense Video Object Captioning (DVOC) is the task of jointly detecting, tracking, and captioning object trajectories in a video, requiring the ability to understand spatio-temporal details and describe them in natural language. Due to the complexity of the task and the high cost associated with manual annotation, previous approaches resort to disjoint training strategies, potentially leading to suboptimal performance. To circumvent this issue, we propose to generate captions about spatio-temporally localized entities leveraging a state-of-the-art VLM. By extending the LVIS and LV-VIS datasets with our synthetic captions (LVISCap and LV-VISCap), we train MaskCaptioner, an end-to-end model capable of jointly detecting, segmenting, tracking and captioning object trajectories. Moreover, with pretraining on LVISCap and LV-VISCap, MaskCaptioner achieves state-of-the-art DVOC results on three existing benchmarks, VidSTG, VLN and BenSMOT. The datasets and code are available at https://www.gabriel.fiastre.fr/maskcaptioner/.

comment: 20 pages, 8 figures

♻ ☆ LatentBreak: Jailbreaking Large Language Models through Latent Space Feedback

Raffaele Mura, Giorgio Piras, Kamilė Lukošiūtė, Maura Pintor, Amin Karbasi, Battista Biggio

Jailbreaks are adversarial attacks designed to bypass the built-in safety mechanisms of large language models. Automated jailbreaks typically optimize an adversarial suffix or adapt long prompt templates by forcing the model to generate the initial part of a restricted or harmful response. In this work, we show that existing jailbreak attacks that leverage such mechanisms to unlock the model response can be detected by a straightforward perplexity-based filtering on the input prompt. To overcome this issue, we propose LatentBreak, a white-box jailbreak attack that generates natural adversarial prompts with low perplexity capable of evading such defenses. LatentBreak substitutes words in the input prompt with semantically-equivalent ones, preserving the initial intent of the prompt, instead of adding high-perplexity adversarial suffixes or long templates. These words are chosen by minimizing the distance in the latent space between the representation of the adversarial prompt and that of harmless requests. Our extensive evaluation shows that LatentBreak leads to shorter and low-perplexity prompts, thus outperforming competing jailbreak algorithms against perplexity-based filters on multiple safety-aligned models.

♻ ☆ SignalLLM: A General-Purpose LLM Agent Framework for Automated Signal Processing

Junlong Ke, Qiying Hu, Shenghai Yuan, Yuecong Xu, Jianfei Yang

Modern signal processing (SP) pipelines, whether model-based or data-driven, often constrained by complex and fragmented workflow, rely heavily on expert knowledge and manual engineering, and struggle with adaptability and generalization under limited data. In contrast, Large Language Models (LLMs) offer strong reasoning capabilities, broad general-purpose knowledge, in-context learning, and cross-modal transfer abilities, positioning them as powerful tools for automating and generalizing SP workflows. Motivated by these potentials, we introduce SignalLLM, the first general-purpose LLM-based agent framework for general SP tasks. Unlike prior LLM-based SP approaches that are limited to narrow applications or tricky prompting, SignalLLM introduces a principled, modular architecture. It decomposes high-level SP goals into structured subtasks via in-context learning and domain-specific retrieval, followed by hierarchical planning through adaptive retrieval-augmented generation (RAG) and refinement; these subtasks are then executed through prompt-based reasoning, cross-modal reasoning, code synthesis, model invocation, or data-driven LLM-assisted modeling. Its generalizable design enables the flexible selection of problem solving strategies across different signal modalities, task types, and data conditions. We demonstrate the versatility and effectiveness of SignalLLM through five representative tasks in communication and sensing, such as radar target detection, human activity recognition, and text compression. Experimental results show superior performance over traditional and existing LLM-based methods, particularly in few-shot and zero-shot settings.

comment: 11 pages

♻ ☆ Differentiation Through Black-Box Quadratic Programming Solvers

Connor W. Magoon, Fengyu Yang, Noam Aigerman, Shahar Z. Kovalsky

Differentiable optimization has attracted significant research interest, particularly for quadratic programming (QP). Existing approaches for differentiating the solution of a QP with respect to its defining parameters often rely on specific integrated solvers. This integration limits their applicability, including their use in neural network architectures and bi-level optimization tasks, restricting users to a narrow selection of solver choices. To address this limitation, we introduce dQP, a modular and solver-agnostic framework for plug-and-play differentiation of virtually any QP solver. A key insight we leverage to achieve modularity is that, once the active set of inequality constraints is known, both the solution and its derivative can be expressed using simplified linear systems that share the same matrix. This formulation fully decouples the computation of the QP solution from its differentiation. Building on this result, we provide a minimal-overhead, open-source implementation ( https://github.com/cwmagoon/dQP ) that seamlessly integrates with over 15 state-of-the-art solvers. Comprehensive benchmark experiments demonstrate dQP's robustness and scalability, particularly highlighting its advantages in large-scale sparse problems.

♻ ☆ Smart Exploration in Reinforcement Learning using Bounded Uncertainty Models

J. S. van Hulst, W. P. M. H. Heemels, D. J. Antunes

Reinforcement learning (RL) is a powerful framework for decision-making in uncertain environments, but it often requires large amounts of data to learn an optimal policy. We address this challenge by incorporating prior model knowledge to guide exploration and accelerate the learning process. Specifically, we assume access to a model set that contains the true transition kernel and reward function. We optimize over this model set to obtain upper and lower bounds on the Q-function, which are then used to guide the exploration of the agent. We provide theoretical guarantees on the convergence of the Q-function to the optimal Q-function under the proposed class of exploring policies. Furthermore, we also introduce a data-driven regularized version of the model set optimization problem that ensures the convergence of the class of exploring policies to the optimal policy. Lastly, we show that when the model set has a specific structure, namely the bounded-parameter MDP (BMDP) framework, the regularized model set optimization problem becomes convex and simple to implement. In this setting, we also prove finite-time convergence to the optimal policy under mild assumptions. We demonstrate the effectiveness of the proposed exploration strategy, which we call BUMEX (Bounded Uncertainty Model-based Exploration), in a simulation study. The results indicate that the proposed method can significantly accelerate learning in benchmark examples. A toolbox is available at https://github.com/JvHulst/BUMEX.

comment: Accepted for Presentation at 64th IEEE Conference on Decision and Control, CDC 2025, Rio de Janeiro, Brazil, 2025

♻ ☆ Epistemic Diversity and Knowledge Collapse in Large Language Models

Dustin Wright, Sarah Masud, Jared Moore, Srishti Yadav, Maria Antoniak, Chan Young Park, Isabelle Augenstein

Large language models (LLMs) tend to generate lexically, semantically, and stylistically homogenous texts. This poses a risk of knowledge collapse, where homogenous LLMs mediate a shrinking in the range of accessible information over time. Existing works on homogenization are limited by a focus on closed-ended multiple-choice setups or fuzzy semantic features, and do not look at trends across time and cultural contexts. To overcome this, we present a new methodology to measure epistemic diversity, i.e., variation in real-world claims in LLM outputs, which we use to perform a broad empirical study of LLM knowledge collapse. We test 27 LLMs, 155 topics covering 12 countries, and 200 prompt variations sourced from real user chats. For the topics in our study, we show that while newer models tend to generate more diverse claims, nearly all models are less epistemically diverse than a basic web search. We find that model size has a negative impact on epistemic diversity, while retrieval-augmented generation (RAG) has a positive impact, though the improvement from RAG varies by the cultural context. Finally, compared to a traditional knowledge source (Wikipedia), we find that country-specific claims reflect the English language more than the local one, highlighting a gap in epistemic representation

comment: 16 pages; 8 figures, 4 tables; v2 changelog: Fixed the modeling for table 3, random effect is the model version; v3 changelog: Fixed minor formatting issues in tables 2 and 3; v4 changelog: Fixed some typos and model description

♻ ☆ Incentivizing LLMs to Self-Verify Their Answers

Fuxiang Zhang, Jiacheng Xu, Chaojie Wang, Ce Cui, Yang Liu, Bo An

Large Language Models (LLMs) have demonstrated remarkable progress in complex reasoning tasks through both post-training and test-time scaling laws. While prevalent test-time scaling approaches are often realized by using external reward models to guide the model generation process, we find that only marginal gains can be acquired when scaling a model post-trained on specific reasoning tasks. We identify that the limited improvement stems from distribution discrepancies between the specific post-trained generator and the general reward model. To address this, we propose a framework that incentivizes LLMs to self-verify their own answers. By unifying answer generation and verification within a single reinforcement learning (RL) process, we train models that can effectively assess the correctness of their own solutions. The trained model can further scale its performance at inference time by verifying its generations, without the need for external verifiers. We train our self-verification models based on Qwen2.5-Math-7B and DeepSeek-R1-Distill-Qwen-1.5B, demonstrating their capabilities across varying reasoning context lengths. Experiments on multiple mathematical reasoning benchmarks show that our models can not only improve post-training performance but also enable effective test-time scaling.

♻ ☆ Human-Like Goalkeeping in a Realistic Football Simulation: a Sample-Efficient Reinforcement Learning Approach

Alessandro Sestini, Joakim Bergdahl, Jean-Philippe Barrette-LaPierre, Florian Fuchs, Brady Chen, Michael Jones, Linus Gisslén

While several high profile video games have served as testbeds for Deep Reinforcement Learning (DRL), this technique has rarely been employed by the game industry for crafting authentic AI behaviors. Previous research focuses on training super-human agents with large models, which is impractical for game studios with limited resources aiming for human-like agents. This paper proposes a sample-efficient DRL method tailored for training and fine-tuning agents in industrial settings such as the video game industry. Our method improves sample efficiency of value-based DRL by leveraging pre-collected data and increasing network plasticity. We evaluate our method training a goalkeeper agent in EA SPORTS FC 25, one of the best-selling football simulations today. Our agent outperforms the game's built-in AI by 10% in ball saving rate. Ablation studies show that our method trains agents 50% faster compared to standard DRL methods. Finally, qualitative evaluation from domain experts indicates that our approach creates more human-like gameplay compared to hand-crafted agents. As a testament to the impact of the approach, the method has been adopted for use in the most recent release of the series.

♻ ☆ A mathematical certification for positivity conditions in Neural Networks with applications to partial monotonicity and Trustworthy AI

Alejandro Polo-Molina, David Alfaya, Jose Portela

Artificial Neural Networks (ANNs) have become a powerful tool for modeling complex relationships in large-scale datasets. However, their black-box nature poses trustworthiness challenges. In certain situations, ensuring trust in predictions might require following specific partial monotonicity constraints. However, certifying if an already-trained ANN is partially monotonic is challenging. Therefore, ANNs are often disregarded in some critical applications, such as credit scoring, where partial monotonicity is required. To address this challenge, this paper presents a novel algorithm (LipVor) that certifies if a black-box model, such as an ANN, is positive based on a finite number of evaluations. Consequently, since partial monotonicity can be expressed as a positivity condition on partial derivatives, LipVor can certify whether an ANN is partially monotonic. To do so, for every positively evaluated point, the Lipschitzianity of the black-box model is used to construct a specific neighborhood where the function remains positive. Next, based on the Voronoi diagram of the evaluated points, a sufficient condition is stated to certify if the function is positive in the domain. Unlike prior methods, our approach certifies partial monotonicity without constrained architectures or piece-wise linear activations. Therefore, LipVor could open up the possibility of using unconstrained ANN in some critical fields. Moreover, some other properties of an ANN, such as convexity, can be posed as positivity conditions, and therefore, LipVor could also be applied.

comment: 16 pages, 4 figures

♻ ☆ GFlowNets for Learning Better Drug-Drug Interaction Representations ICANN 2025

Azmine Toushik Wasi

Drug-drug interactions pose a significant challenge in clinical pharmacology, with severe class imbalance among interaction types limiting the effectiveness of predictive models. Common interactions dominate datasets, while rare but critical interactions remain underrepresented, leading to poor model performance on infrequent cases. Existing methods often treat DDI prediction as a binary problem, ignoring class-specific nuances and exacerbating bias toward frequent interactions. To address this, we propose a framework combining Generative Flow Networks (GFlowNet) with Variational Graph Autoencoders (VGAE) to generate synthetic samples for rare classes, improving model balance and generate effective and novel DDI pairs. Our approach enhances predictive performance across interaction types, ensuring better clinical reliability.

comment: Accepted to ICANN 2025:AIDD and NeurIPS 2025 Workshop on Structured Probabilistic Inference & Generative Modeling (https://openreview.net/forum?id=LZW1jSgfCI)

♻ ☆ Exploring the Early Universe with Deep Learning

Emmanuel de Salis, Massimo De Santis, Davide Piras, Sambit K. Giri, Michele Bianco, Nicolas Cerardi, Philipp Denzel, Merve Selcuk-Simsek, Kelley M. Hess, M. Carmen Toribio, Franz Kirsten, Hatem Ghorbel

Hydrogen is the most abundant element in our Universe. The first generation of stars and galaxies produced photons that ionized hydrogen gas, driving a cosmological event known as the Epoch of Reionization (EoR). The upcoming Square Kilometre Array Observatory (SKAO) will map the distribution of neutral hydrogen during this era, aiding in the study of the properties of these first-generation objects. Extracting astrophysical information will be challenging, as SKAO will produce a tremendous amount of data where the hydrogen signal will be contaminated with undesired foreground contamination and instrumental systematics. To address this, we develop the latest deep learning techniques to extract information from the 2D power spectra of the hydrogen signal expected from SKAO. We apply a series of neural network models to these measurements and quantify their ability to predict the history of cosmic hydrogen reionization, which is connected to the increasing number and efficiency of early photon sources. We show that the study of the early Universe benefits from modern deep learning technology. In particular, we demonstrate that dedicated machine learning algorithms can achieve more than a $0.95$ $R^2$ score on average in recovering the reionization history. This enables accurate and precise cosmological and astrophysical inference of structure formation in the early Universe.

comment: EPIA 2025 preprint version, 12 pages, 3 figures

♻ ☆ Rethinking Optimal Verification Granularity for Compute-Efficient Test-Time Scaling NeurIPS 2025

Hao Mark Chen, Guanxi Lu, Yasuyuki Okoshi, Zhiwen Mo, Masato Motomura, Hongxiang Fan

Test-time scaling (TTS) has proven effective in enhancing the reasoning capabilities of large language models (LLMs). Verification plays a key role in TTS, simultaneously influencing (1) reasoning performance and (2) compute efficiency, due to the quality and computational cost of verification. In this work, we challenge the conventional paradigms of verification, and make the first attempt toward systematically investigating the impact of verification granularity-that is, how frequently the verifier is invoked during generation, beyond verifying only the final output or individual generation steps. To this end, we introduce Variable Granularity Search (VG-Search), a unified algorithm that generalizes beam search and Best-of-N sampling via a tunable granularity parameter g. Extensive experiments with VG-Search under varying compute budgets, generator-verifier configurations, and task attributes reveal that dynamically selecting g can improve the compute efficiency and scaling behavior. Building on these findings, we propose adaptive VG-Search strategies that achieve accuracy gains of up to 3.1\% over Beam Search and 3.6\% over Best-of-N, while reducing FLOPs by over 52\%. We will open-source the code to support future research.

comment: Accepted at NeurIPS 2025

♻ ☆ A geometric framework for momentum-based optimizers for low-rank training

Steffen Schotthöfer, Timon Klein, Jonas Kusch

Low-rank pre-training and fine-tuning have recently emerged as promising techniques for reducing the computational and storage costs of large neural networks. Training low-rank parameterizations typically relies on conventional optimizers such as heavy ball momentum methods or Adam. In this work, we identify and analyze potential difficulties that these training methods encounter when used to train low-rank parameterizations of weights. In particular, we show that classical momentum methods can struggle to converge to a local optimum due to the geometry of the underlying optimization landscape. To address this, we introduce novel training strategies derived from dynamical low-rank approximation, which explicitly account for the underlying geometric structure. Our approach leverages and combines tools from dynamical low-rank approximation and momentum-based optimization to design optimizers that respect the intrinsic geometry of the parameter space. We validate our methods through numerical experiments, demonstrating faster convergence, and stronger validation metrics at given parameter budgets.

♻ ☆ LLM-Driven Treatment Effect Estimation Under Inference Time Text Confounding

Yuchen Ma, Dennis Frauen, Jonas Schweisthal, Stefan Feuerriegel

Estimating treatment effects is crucial for personalized decision-making in medicine, but this task faces unique challenges in clinical practice. At training time, models for estimating treatment effects are typically trained on well-structured medical datasets that contain detailed patient information. However, at inference time, predictions are often made using textual descriptions (e.g., descriptions with self-reported symptoms), which are incomplete representations of the original patient information. In this work, we make three contributions. (1) We show that the discrepancy between the data available during training time and inference time can lead to biased estimates of treatment effects. We formalize this issue as an inference time text confounding problem, where confounders are fully observed during training time but only partially available through text at inference time. (2) To address this problem, we propose a novel framework for estimating treatment effects that explicitly accounts for inference time text confounding. Our framework leverages large language models together with a custom doubly robust learner to mitigate biases caused by the inference time text confounding. (3) Through a series of experiments, we demonstrate the effectiveness of our framework in real-world applications.

♻ ☆ MedAgentBoard: Benchmarking Multi-Agent Collaboration with Conventional Methods for Diverse Medical Tasks NeurIPS 2025

Yinghao Zhu, Ziyi He, Haoran Hu, Xiaochen Zheng, Xichen Zhang, Zixiang Wang, Junyi Gao, Liantao Ma, Lequan Yu

The rapid advancement of Large Language Models (LLMs) has stimulated interest in multi-agent collaboration for addressing complex medical tasks. However, the practical advantages of multi-agent collaboration approaches remain insufficiently understood. Existing evaluations often lack generalizability, failing to cover diverse tasks reflective of real-world clinical practice, and frequently omit rigorous comparisons against both single-LLM-based and established conventional methods. To address this critical gap, we introduce MedAgentBoard, a comprehensive benchmark for the systematic evaluation of multi-agent collaboration, single-LLM, and conventional approaches. MedAgentBoard encompasses four diverse medical task categories: (1) medical (visual) question answering, (2) lay summary generation, (3) structured Electronic Health Record (EHR) predictive modeling, and (4) clinical workflow automation, across text, medical images, and structured EHR data. Our extensive experiments reveal a nuanced landscape: while multi-agent collaboration demonstrates benefits in specific scenarios, such as enhancing task completeness in clinical workflow automation, it does not consistently outperform advanced single LLMs (e.g., in textual medical QA) or, critically, specialized conventional methods that generally maintain better performance in tasks like medical VQA and EHR-based prediction. MedAgentBoard offers a vital resource and actionable insights, emphasizing the necessity of a task-specific, evidence-based approach to selecting and developing AI solutions in medicine. It underscores that the inherent complexity and overhead of multi-agent collaboration must be carefully weighed against tangible performance gains. All code, datasets, detailed prompts, and experimental results are open-sourced at https://medagentboard.netlify.app/.

comment: Accepted by NeurIPS 2025 Datasets & Benchmarks Track

♻ ☆ Distributed optimization: designed for federated learning

Wenyou Guo, Ting Qu, Chunrong Pan, George Q. Huang

Federated learning (FL), as a distributed collaborative machine learning (ML) framework under privacy-preserving constraints, has garnered increasing research attention in cross-organizational data collaboration scenarios. This paper proposes a class of distributed optimization algorithms based on the augmented Lagrangian technique, designed to accommodate diverse communication topologies in both centralized and decentralized FL settings. Furthermore, we develop multiple termination criteria and parameter update mechanisms to enhance computational efficiency, accompanied by rigorous theoretical guarantees of convergence. By generalizing the augmented Lagrangian relaxation through the incorporation of proximal relaxation and quadratic approximation, our framework systematically recovers a broad of classical unconstrained optimization methods, including proximal algorithm, classic gradient descent, and stochastic gradient descent, among others. Notably, the convergence properties of these methods can be naturally derived within the proposed theoretical framework. Numerical experiments demonstrate that the proposed algorithm exhibits strong performance in large-scale settings with significant statistical heterogeneity across clients.

comment: 16 pages, 6 figures

♻ ☆ On the Impact of Performative Risk Minimization for Binary Random Variables ICML 2025

Nikita Tsoy, Ivan Kirev, Negin Rahimiyazdi, Nikola Konstantinov

Performativity, the phenomenon where outcomes are influenced by predictions, is particularly prevalent in social contexts where individuals strategically respond to a deployed model. In order to preserve the high accuracy of machine learning models under distribution shifts caused by performativity, Perdomo et al. (2020) introduced the concept of performative risk minimization (PRM). While this framework ensures model accuracy, it overlooks the impact of the PRM on the underlying distributions and the predictions of the model. In this paper, we initiate the analysis of the impact of PRM, by studying performativity for a sequential performative risk minimization problem with binary random variables and linear performative shifts. We formulate two natural measures of impact. In the case of full information, where the distribution dynamics are known, we derive explicit formulas for the PRM solution and our impact measures. In the case of partial information, we provide performative-aware statistical estimators, as well as simulations. Our analysis contrasts PRM to alternatives that do not model data shift and indicates that PRM can have amplified side effects compared to such methods.

comment: ICML 2025 camera-ready

♻ ☆ Machine-learning competition to grade EEG background patterns in newborns with hypoxic-ischaemic encephalopathy

Fabio Magarelli, Geraldine B. Boylan, Saeed Montazeri, Feargal O'Sullivan, Dominic Lightbody, Minoo Ashoori, Tamara Skoric, John M. O'Toole

Machine learning (ML) has the potential to support and improve expert performance in monitoring the brain function of at-risk newborns. Developing accurate and reliable ML models depends on access to high-quality, annotated data, a resource in short supply. ML competitions address this need by providing researchers access to expertly annotated datasets, fostering shared learning through direct model comparisons, and leveraging the benefits of crowdsourcing diverse expertise. We compiled a retrospective dataset containing 353 hours of EEG from 102 individual newborns from a multi-centre study. The data was fully anonymised and divided into training, testing, and held-out validation datasets. EEGs were graded for the severity of abnormal background patterns. Next, we created a web-based competition platform and hosted a machine learning competition to develop ML models for classifying the severity of EEG background patterns in newborns. After the competition closed, the top 4 performing models were evaluated offline on a separate held-out validation dataset. Although a feature-based model ranked first on the testing dataset, deep learning models generalised better on the validation sets. All methods had a significant decline in validation performance compared to the testing performance. This highlights the challenges for model generalisation on unseen data, emphasising the need for held-out validation datasets in ML studies with neonatal EEG. The study underscores the importance of training ML models on large and diverse datasets to ensure robust generalisation. The competition's outcome demonstrates the potential for open-access data and collaborative ML development to foster a collaborative research environment and expedite the development of clinical decision-support tools for neonatal neuromonitoring.

comment: 29 pages, supplementary materials: "supplementary materials ML Comp.docx"

♻ ☆ Wisdom and Delusion of LLM Ensembles for Code Generation and Repair

Fernando Vallecillos-Ruiz, Max Hort, Leon Moonen

Today's pursuit of a single Large Language Model (LMM) for all software engineering tasks is resource-intensive and overlooks the potential benefits of complementarity, where different models contribute unique strengths. However, the degree to which coding LLMs complement each other and the best strategy for maximizing an ensemble's potential are unclear, leaving practitioners without a clear path to move beyond single-model systems. To address this gap, we empirically compare ten individual LLMs from five families, and three ensembles of these LLMs across three software engineering benchmarks covering code generation and program repair. We assess the complementarity between models and the performance gap between the best individual model and the ensembles. Next, we evaluate various selection heuristics to identify correct solutions from an ensemble's candidate pool. We find that the theoretical upperbound for an ensemble's performance can be 83% above the best single model. Our results show that consensus-based strategies for selecting solutions fall into a "popularity trap," amplifying common but incorrect outputs. In contrast, a diversity-based strategy realizes up to 95% of this theoretical potential, and proves effective even in small two-model ensembles, enabling a cost-efficient way to enhance performance by leveraging multiple LLMs.

comment: Added Acknowledgments section and hyphenated last names

♻ ☆ Cybersecurity threat detection based on a UEBA framework using Deep Autoencoders

Jose Fuentes, Ines Ortega-Fernandez, Nora M. Villanueva, Marta Sestelo

User and Entity Behaviour Analytics (UEBA) is a broad branch of data analytics that attempts to build a normal behavioural profile in order to detect anomalous events. Among the techniques used to detect anomalies, Deep Autoencoders constitute one of the most promising deep learning models on UEBA tasks, allowing explainable detection of security incidents that could lead to the leak of personal data, hijacking of systems, or access to sensitive business information. In this study, we introduce the first implementation of an explainable UEBA-based anomaly detection framework that leverages Deep Autoencoders in combination with Doc2Vec to process both numerical and textual features. Additionally, based on the theoretical foundations of neural networks, we offer a novel proof demonstrating the equivalence of two widely used definitions for fully-connected neural networks. The experimental results demonstrate the proposed framework capability to detect real and synthetic anomalies effectively generated from real attack data, showing that the models provide not only correct identification of anomalies but also explainable results that enable the reconstruction of the possible origin of the anomaly. Our findings suggest that the proposed UEBA framework can be seamlessly integrated into enterprise environments, complementing existing security systems for explainable threat detection.

comment: Published in AIMS Mathematics (2025), 10(10): 23496-23517. DOI: 10.3934/math.20251043

♻ ☆ Monopoly Deal: A Benchmark Environment for Bounded One-Sided Response Games

Will Wolf

Card games are widely used to study sequential decision-making under uncertainty, with real-world analogues in negotiation, finance, and cybersecurity. These games typically fall into three categories based on the flow of control: strictly sequential (players alternate single actions), deterministic response (some actions trigger a fixed outcome), and unbounded reciprocal response (alternating counterplays are permitted). A less-explored but strategically rich structure is the bounded one-sided response, where a player's action briefly transfers control to the opponent, who must satisfy a fixed condition through one or more moves before the turn resolves. We term games featuring this mechanism Bounded One-Sided Response Games (BORGs). We introduce a modified version of Monopoly Deal as a benchmark environment that isolates this dynamic, where a Rent action forces the opponent to choose payment assets. The gold-standard algorithm, Counterfactual Regret Minimization (CFR), converges on effective strategies without novel algorithmic extensions. A lightweight full-stack research platform unifies the environment, a parallelized CFR runtime, and a human-playable web interface. The trained CFR agent and source code are available at https://monopolydeal.ai.

comment: 24 pages, 7 figures

♻ ☆ In Defence of Post-hoc Explainability NeurIPS 2024

Nick Oh

This position paper defends post-hoc explainability methods as legitimate tools for scientific knowledge production in machine learning. Addressing criticism of these methods' reliability and epistemic status, we develop a philosophical framework grounded in mediated understanding and bounded factivity. We argue that scientific insights can emerge through structured interpretation of model behaviour without requiring complete mechanistic transparency, provided explanations acknowledge their approximative nature and undergo rigorous empirical validation. Through analysis of recent biomedical ML applications, we demonstrate how post-hoc methods, when properly integrated into scientific practice, generate novel hypotheses and advance phenomenal understanding.

comment: v1 presented at the Interpretable AI: Past, Present, and Future Workshop at NeurIPS 2024 (non-archival)

♻ ☆ On-the-Fly OVD Adaptation with FLAME: Few-shot Localization via Active Marginal-Samples Exploration

Yehonathan Refael, Amit Aides, Aviad Barzilai, George Leifman, Genady Beryozkin, Vered Silverman, Bolous Jaber, Tomer Shekel

Open-vocabulary object detection (OVD) models offer remarkable flexibility by detecting objects from arbitrary text queries. However, their zero-shot performance in specialized domains like Remote Sensing (RS) is often compromised by the inherent ambiguity of natural language, limiting critical downstream applications. For instance, an OVD model may struggle to distinguish between fine-grained classes such as "fishing boat" and "yacht" since their embeddings are similar and often inseparable. This can hamper specific user goals, such as monitoring illegal fishing, by producing irrelevant detections. To address this, we propose a cascaded approach that couples the broad generalization of a large pre-trained OVD model with a lightweight few-shot classifier. Our method first employs the zero-shot model to generate high-recall object proposals. These proposals are then refined for high precision by a compact classifier trained in real-time on only a handful of user-annotated examples - drastically reducing the high costs of RS imagery annotation.The core of our framework is FLAME, a one-step active learning strategy that selects the most informative samples for training. FLAME identifies, on the fly, uncertain marginal candidates near the decision boundary using density estimation, followed by clustering to ensure sample diversity. This efficient sampling technique achieves high accuracy without costly full-model fine-tuning and enables instant adaptation, within less then a minute, which is significantly faster than state-of-the-art alternatives.Our method consistently surpasses state-of-the-art performance on RS benchmarks, establishing a practical and resource-efficient framework for adapting foundation models to specific user needs.

♻ ☆ Seeing Structural Failure Before it Happens: An Image-Based Physics-Informed Neural Network (PINN) for Spaghetti Bridge Load Prediction

Omer Jauhar Khan, Sudais Khan, Hafeez Anwar, Shahzeb Khan, Shams Ul Arifeen

Physics Informed Neural Networks (PINNs) are gaining attention for their ability to embed physical laws into deep learning models, which is particularly useful in structural engineering tasks with limited data. This paper aims to explore the use of PINNs to predict the weight of small scale spaghetti bridges, a task relevant to understanding load limits and potential failure modes in simplified structural models. Our proposed framework incorporates physics-based constraints to the prediction model for improved performance. In addition to standard PINNs, we introduce a novel architecture named Physics Informed Kolmogorov Arnold Network (PIKAN), which blends universal function approximation theory with physical insights. The structural parameters provided as input to the model are collected either manually or through computer vision methods. Our dataset includes 15 real bridges, augmented to 100 samples, and our best model achieves an $R^2$ score of 0.9603 and a mean absolute error (MAE) of 10.50 units. From applied perspective, we also provide a web based interface for parameter entry and prediction. These results show that PINNs can offer reliable estimates of structural weight, even with limited data, and may help inform early stage failure analysis in lightweight bridge designs. The complete data and code are available at https://github.com/OmerJauhar/PINNS-For-Spaghetti-Bridges.

comment: 12 pages, 17 figures. Preprint

♻ ☆ TokenWeave: Efficient Compute-Communication Overlap for Distributed LLM Inference

Raja Gond, Nipun Kwatra, Ramachandran Ramjee

Distributed inference of large language models (LLMs) can introduce overheads of up to 20% even over GPUs connected via high-speed interconnects such as NVLink. Multiple techniques have been proposed to mitigate these overheads by decomposing computations into finer-grained tasks and overlapping communication with sub-tasks as they complete. However, fine-grained decomposition of a large computation into many smaller computations on GPUs results in overheads. Furthermore, the communication itself uses many streaming multiprocessors (SMs), adding to the overhead. We present TokenWeave to address these challenges. TokenWeave proposes a Token-Splitting technique that divides the tokens in the inference batch into two approximately equal subsets in a wave-aware manner. The communication of one subset is then overlapped with the computation of the other. In addition, TokenWeave optimizes the order of the layer normalization computation with respect to communication operations and implements a novel fused AllReduce--RMSNorm kernel that carefully leverages Multimem instruction support available on Hopper and Blackwell NVIDIA GPUs. These optimizations allow TokenWeave to perform communication and RMSNorm using only 2-8 SMs. Moreover, our kernel enables the memory-bound RMSNorm to be overlapped with the other batch's computation, providing additional gains. Our evaluations demonstrate up to 1.29x speedup in latency and 1.26x higher throughput across multiple models and workloads. In several settings, TokenWeave results in better performance compared to an equivalent model with all communication removed.

comment: 14 pages, 16 figures. For source code, see https://github.com/microsoft/tokenweave. In version 2, Figure 6 shows All-Reduce bandwidth instead of Reduce-Scatter. The Multimem Reduce-Scatter bandwidth formula differs slightly from the ring-based version. Fixed x-ticks in Figure 7

♻ ☆ Nearly Minimax Discrete Distribution Estimation in Kullback-Leibler Divergence with High Probability

Dirk van der Hoeven, Julia Olkhovskaia, Tim van Erven

We consider the fundamental problem of estimating a discrete distribution on a domain of size~$K$ with high probability in Kullback-Leibler divergence. We provide upper and lower bounds on the minimax estimation rate, which show that the optimal rate is between $\big(K + \ln(K)\ln(1/\delta)\big) /n$ and $\big(K\ln\ln(K) + \ln(K)\ln(1/\delta)\big) /n$ at error probability $\delta$ and sample size $n$, which pins down the rate up to the doubly logarithmic factor $\ln \ln K$ that multiplies $K$. Our upper bound uses techniques from online learning to construct a novel estimator via online-to-batch conversion. Perhaps surprisingly, the tail behavior of the minimax rate is worse than for the squared total variation and squared Hellinger distance, for which it is $\big(K + \ln(1/\delta)\big) /n$, i.e.\ without the $\ln K$ multiplying $\ln (1/\delta)$. As a consequence, we cannot obtain a fully tight lower bound from the usual reduction to these smaller distances. Moreover, we show that this lower bound cannot be achieved by the standard lower bound approach based on a reduction to hypothesis testing, and instead we need to introduce a new reduction to what we call weak hypothesis testing. We investigate the source of the gap with other divergences further in refined results, which show that the total variation rate is achievable for Kullback-Leibler divergence after all (in fact by he maximum likelihood estimator) if we rule out outcome probabilities smaller than $O(\ln(K/\delta) / n)$, which is a vanishing set as $n$ increases for fixed $K$ and~$\delta$. This explains why minimax Kullback-Leibler estimation is more difficult than asymptotic estimation.

♻ ☆ Predictive Causal Inference via Spatio-Temporal Modeling and Penalized Empirical Likelihood

Byunghee Lee, Hye Yeon Sin, Joonsung Kang

This study introduces an integrated framework for predictive causal inference designed to overcome limitations inherent in conventional single model approaches. Specifically, we combine a Hidden Markov Model (HMM) for spatial health state estimation with a Multi Task and Multi Graph Convolutional Network (MTGCN) for capturing temporal outcome trajectories. The framework asymmetrically treats temporal and spatial information regarding them as endogenous variables in the outcome regression, and exogenous variables in the propensity score model, thereby expanding the standard doubly robust treatment effect estimation to jointly enhance bias correction and predictive accuracy. To demonstrate its utility, we focus on clinical domains such as cancer, dementia, and Parkinson disease, where treatment effects are challenging to observe directly. Simulation studies are conducted to emulate latent disease dynamics and evaluate the model performance under varying conditions. Overall, the proposed framework advances predictive causal inference by structurally adapting to spatiotemporal complexities common in biomedical data.

♻ ☆ Towards a Method for Synthetic Generation of Persons with Aphasia Transcripts

Jason M. Pittman, Anton Phillips Jr., Yesenia Medina-Santos, Brielle C. Stark

In aphasia research, Speech-Language Pathologists (SLPs) devote extensive time to manually coding speech samples using Correct Information Units (CIUs), a measure of how informative an individual sample of speech is. Developing automated systems to recognize aphasic language is limited by data scarcity. For example, only about 600 transcripts are available in AphasiaBank yet billions of tokens are used to train large language models (LLMs). In the broader field of machine learning (ML), researchers increasingly turn to synthetic data when such are sparse. Therefore, this study constructs and validates two methods to generate synthetic transcripts of the AphasiaBank Cat Rescue picture description task. One method leverages a procedural programming approach while the second uses Mistral 7b Instruct and Llama 3.1 8b Instruct LLMs. The methods generate transcripts across four severity levels (Mild, Moderate, Severe, Very Severe) through word dropping, filler insertion, and paraphasia substitution. Overall, we found, compared to human-elicited transcripts, Mistral 7b Instruct best captures key aspects of linguistic degradation observed in aphasia, showing realistic directional changes in NDW, word count, and word length amongst the synthetic generation methods. Based on the results, future work should plan to create a larger dataset, fine-tune models for better aphasic representation, and have SLPs assess the realism and usefulness of the synthetic transcripts.

comment: 19 pages, 1 figure, 7 tables

♻ ☆ SafEDMD: A Koopman-based data-driven controller design framework for nonlinear dynamical systems

Robin Strässer, Manuel Schaller, Karl Worthmann, Julian Berberich, Frank Allgöwer

The Koopman operator serves as the theoretical backbone for machine learning of dynamical control systems, where the operator is heuristically approximated by extended dynamic mode decomposition (EDMD). In this paper, we propose SafEDMD, a novel stability- and feedback-oriented EDMD-based controller design framework. Our approach leverages a reliable surrogate model generated in a data-driven fashion in order to provide closed-loop guarantees. In particular, we establish a controller design based on semi-definite programming with guaranteed stabilization of the underlying nonlinear system. As central ingredient, we derive proportional error bounds that vanish at the origin and are tailored to control tasks. We illustrate the developed method by means of several benchmark examples and highlight the advantages over state-of-the-art methods.

comment: Accepted for publication in Automatica

♻ ☆ Mind the Gap: Removing the Discretization Gap in Differentiable Logic Gate Networks NeurIPS 2025

Shakir Yousefi, Andreas Plesner, Till Aczel, Roger Wattenhofer

Modern neural networks demonstrate state-of-the-art performance on numerous existing benchmarks; however, their high computational requirements and energy consumption prompt researchers to seek more efficient solutions for real-world deployment. Logic gate networks (LGNs) learns a large network of logic gates for efficient image classification. However, learning a network that can solve a simple problem like CIFAR-10 can take days to weeks to train. Even then, almost half of the network remains unused, causing a discretization gap. This discretization gap hinders real-world deployment of LGNs, as the performance drop between training and inference negatively impacts accuracy. We inject Gumbel noise with a straight-through estimator during training to significantly speed up training, improve neuron utilization, and decrease the discretization gap. We theoretically show that this results from implicit Hessian regularization, which improves the convergence properties of LGNs. We train networks $4.5 \times$ faster in wall-clock time, reduce the discretization gap by $98\%$, and reduce the number of unused gates by $100\%$.

comment: Accepted to NeurIPS 2025 (main track)

♻ ☆ Elementary, My Dear Watson: Non-Invasive Neural Keyword Spotting in the LibriBrain Dataset

Gereon Elvers, Gilad Landau, Oiwi Parker Jones

Non-invasive brain-computer interfaces (BCIs) are beginning to benefit from large, public benchmarks. However, current benchmarks target relatively simple, foundational tasks like Speech Detection and Phoneme Classification, while application-ready results on tasks like Brain-to-Text remain elusive. We propose Keyword Spotting (KWS) as a practically applicable, privacy-aware intermediate task. Using the deep 52-hour, within-subject LibriBrain corpus, we provide standardized train/validation/test splits for reproducible benchmarking, and adopt an evaluation protocol tailored to extreme class imbalance. Concretely, we use area under the precision-recall curve (AUPRC) as a robust evaluation metric, complemented by false alarms per hour (FA/h) at fixed recall to capture user-facing trade-offs. To simplify deployment and further experimentation within the research community, we are releasing an updated version of the pnpl library with word-level dataloaders and Colab-ready tutorials. As an initial reference model, we present a compact 1-D Conv/ResNet baseline with focal loss and top-k pooling that is trainable on a single consumer-class GPU. The reference model achieves approximately 13x the permutation baseline AUPRC on held-out sessions, demonstrating the viability of the task. Exploratory analyses reveal: (i) predictable within-subject scaling - performance improves log-linearly with more training hours - and (ii) the existence of word-level factors (frequency and duration) that systematically modulate detectability.

comment: 16 pages, 7 figures, 6 tables; updated acknowledgments

♻ ☆ AIMeter: Measuring, Analyzing, and Visualizing Energy and Carbon Footprint of AI Workloads

Hongzhen Huang, Kunming Zhang, Hanlong Liao, Kui Wu, Guoming Tang

The rapid advancement of AI, particularly large language models (LLMs), has raised significant concerns about the energy use and carbon emissions associated with model training and inference. However, existing tools for measuring and reporting such impacts are often fragmented, lacking systematic metric integration and offering limited support for correlation analysis among them. This paper presents AIMeter, a comprehensive software toolkit for the measurement, analysis, and visualization of energy use, power draw, hardware performance, and carbon emissions across AI workloads. By seamlessly integrating with existing AI frameworks, AIMeter offers standardized reports and exports fine-grained time-series data to support benchmarking and reproducibility in a lightweight manner. It further enables in-depth correlation analysis between hardware metrics and model performance and thus facilitates bottleneck identification and performance enhancement. By addressing critical limitations in existing tools, AIMeter encourages the research community to weigh environmental impact alongside raw performance of AI workloads and advances the shift toward more sustainable "Green AI" practices. The code is available at https://github.com/SusCom-Lab/AIMeter.

comment: 11 pages, 7 figures and 5 tables

♻ ☆ Stability and Sharper Risk Bounds with Convergence Rate $\tilde{O}(1/n^2)$

Bowei Zhu, Shaojie Li, Mingyang Yi, Yong Liu

Prior work (Klochkov $\&$ Zhivotovskiy, 2021) establishes at most $O\left(\log (n)/n\right)$ excess risk bounds via algorithmic stability for strongly-convex learners with high probability. We show that under the similar common assumptions -- - Polyak-Lojasiewicz condition, smoothness, and Lipschitz continous for losses -- - rates of $O\left(\log^2(n)/n^2\right)$ are at most achievable. To our knowledge, our analysis also provides the tightest high-probability bounds for gradient-based generalization gaps in nonconvex settings.

♻ ☆ Hysteresis Activation Function for Efficient Inference NeurIPS

Moshe Kimhi, Idan Kashani, Avi Mendelson, Chaim Baskin

The widely used ReLU is favored for its hardware efficiency, {as the implementation at inference is a one bit sign case,} yet suffers from issues such as the ``dying ReLU'' problem, where during training, neurons fail to activate and constantly remain at zero, as highlighted by Lu et al. Traditional approaches to mitigate this issue often introduce more complex and less hardware-friendly activation functions. In this work, we propose a Hysteresis Rectified Linear Unit (HeLU), an efficient activation function designed to address the ``dying ReLU'' problem with minimal complexity. Unlike traditional activation functions with fixed thresholds for training and inference, HeLU employs a variable threshold that refines the backpropagation. This refined mechanism allows simpler activation functions to achieve competitive performance comparable to their more complex counterparts without introducing unnecessary complexity or requiring inductive biases. Empirical evaluations demonstrate that HeLU enhances model generalization across diverse datasets, offering a promising solution for efficient and effective inference suitable for a wide range of neural network architectures.

comment: Accepted to 4th NeurIPS Efficient Natural Language and Speech Processing Workshop (ENLSP-IV 2024)

♻ ☆ GRPO-Guard: Mitigating Implicit Over-Optimization in Flow Matching via Regulated Clipping

Jing Wang, Jiajun Liang, Jie Liu, Henglin Liu, Gongye Liu, Jun Zheng, Wanyuan Pang, Ao Ma, Zhenyu Xie, Xintao Wang, Meng Wang, Pengfei Wan, Xiaodan Liang

Recently, GRPO-based reinforcement learning has shown remarkable progress in optimizing flow-matching models, effectively improving their alignment with task-specific rewards. Within these frameworks, the policy update relies on importance-ratio clipping to constrain overconfident positive and negative gradients. However, in practice, we observe a systematic shift in the importance-ratio distribution-its mean falls below 1 and its variance differs substantially across timesteps. This left-shifted and inconsistent distribution prevents positive-advantage samples from entering the clipped region, causing the mechanism to fail in constraining overconfident positive updates. As a result, the policy model inevitably enters an implicit over-optimization stage-while the proxy reward continues to increase, essential metrics such as image quality and text-prompt alignment deteriorate sharply, ultimately making the learned policy impractical for real-world use. To address this issue, we introduce GRPO-Guard, a simple yet effective enhancement to existing GRPO frameworks. Our method incorporates ratio normalization, which restores a balanced and step-consistent importance ratio, ensuring that PPO clipping properly constrains harmful updates across denoising timesteps. In addition, a gradient reweighting strategy equalizes policy gradients over noise conditions, preventing excessive updates from particular timestep regions. Together, these designs act as a regulated clipping mechanism, stabilizing optimization and substantially mitigating implicit over-optimization without relying on heavy KL regularization. Extensive experiments on multiple diffusion backbones (e.g., SD3.5M, Flux.1-dev) and diverse proxy tasks demonstrate that GRPO-Guard significantly reduces over-optimization while maintaining or even improving generation quality.

comment: Project Page: https://jingw193.github.io/GRPO-Guard/

♻ ☆ Oryx: a Scalable Sequence Model for Many-Agent Coordination in Offline MARL NeurIPS 2025

Claude Formanek, Omayma Mahjoub, Louay Ben Nessir, Sasha Abramowitz, Ruan de Kock, Wiem Khlifi, Daniel Rajaonarivonivelomanantsoa, Simon Du Toit, Arnol Fokam, Siddarth Singh, Ulrich Mbou Sob, Felix Chalumeau, Arnu Pretorius

A key challenge in offline multi-agent reinforcement learning (MARL) is achieving effective many-agent multi-step coordination in complex environments. In this work, we propose Oryx, a novel algorithm for offline cooperative MARL to directly address this challenge. Oryx adapts the recently proposed retention-based architecture Sable and combines it with a sequential form of implicit constraint Q-learning (ICQ), to develop a novel offline autoregressive policy update scheme. This allows Oryx to solve complex coordination challenges while maintaining temporal coherence over long trajectories. We evaluate Oryx across a diverse set of benchmarks from prior works -- SMAC, RWARE, and Multi-Agent MuJoCo -- covering tasks of both discrete and continuous control, varying in scale and difficulty. Oryx achieves state-of-the-art performance on more than 80% of the 65 tested datasets, outperforming prior offline MARL methods and demonstrating robust generalisation across domains with many agents and long horizons. Finally, we introduce new datasets to push the limits of many-agent coordination in offline MARL, and demonstrate Oryx's superior ability to scale effectively in such settings.

comment: Published at the 39th Conference on Neural Information Processing Systems (NeurIPS 2025)

♻ ☆ Neurosymbolic Diffusion Models NeurIPS 2025

Emile van Krieken, Pasquale Minervini, Edoardo Ponti, Antonio Vergari

Neurosymbolic (NeSy) predictors combine neural perception with symbolic reasoning to solve tasks like visual reasoning. However, standard NeSy predictors assume conditional independence between the symbols they extract, thus limiting their ability to model interactions and uncertainty - often leading to overconfident predictions and poor out-of-distribution generalisation. To overcome the limitations of the independence assumption, we introduce neurosymbolic diffusion models (NeSyDMs), a new class of NeSy predictors that use discrete diffusion to model dependencies between symbols. Our approach reuses the independence assumption from NeSy predictors at each step of the diffusion process, enabling scalable learning while capturing symbol dependencies and uncertainty quantification. Across both synthetic and real-world benchmarks - including high-dimensional visual path planning and rule-based autonomous driving - NeSyDMs achieve state-of-the-art accuracy among NeSy predictors and demonstrate strong calibration.

comment: Accepted to NeurIPS 2025

♻ ☆ AnomalyMatch: Discovering Rare Objects of Interest with Semi-supervised and Active Learning

Pablo Gómez, Laslo E. Ruhberg, Maria Teresa Nardone, David O'Ryan

Anomaly detection in large datasets is essential in astronomy and computer vision. However, due to a scarcity of labelled data, it is often infeasible to apply supervised methods to anomaly detection. We present AnomalyMatch, an anomaly detection framework combining the semi-supervised FixMatch algorithm using EfficientNet classifiers with active learning. AnomalyMatch is tailored for large-scale applications and integrated into the ESA Datalabs science platform. In this method, we treat anomaly detection as a binary classification problem and efficiently utilise limited labelled and abundant unlabelled images for training. We enable active learning via a user interface for verification of high-confidence anomalies and correction of false positives. Evaluations on the GalaxyMNIST astronomical dataset and the miniImageNet natural-image benchmark under severe class imbalance display strong performance. Starting from five to ten labelled anomalies, we achieve an average AUROC of 0.96 (miniImageNet) and 0.89 (GalaxyMNIST), with respective AUPRC of 0.82 and 0.77. After three active learning cycles, anomalies are ranked with 76% (miniImageNet) to 94% (GalaxyMNIST) precision in the top 1% of the highest-ranking images by score. We compare to the established Astronomaly software on selected 'odd' galaxies from the 'Galaxy Zoo - The Galaxy Challenge' dataset, achieving comparable performance with an average AUROC of 0.83. Our results underscore the exceptional utility and scalability of this approach for anomaly discovery, highlighting the value of specialised approaches for domains characterised by severe label scarcity.

comment: Journal submission in preparation to RASTI; 15 pages; 12 figures

♻ ☆ Diversity as a Reward: Fine-Tuning LLMs on a Mixture of Domain-Undetermined Data NeurIPS'25

Zhenqing Ling, Daoyuan Chen, Liuyi Yao, Qianli Shen, Yaliang Li, Ying Shen

Fine-tuning large language models (LLMs) using diverse datasets is crucial for enhancing their overall performance across various domains. In practical scenarios, existing methods based on modeling the mixture proportions of data composition often struggle with data whose domain labels are missing, imprecise or non-normalized, while methods based on data selection usually encounter difficulties in balancing multi-domain performance. To address these challenges, in this work, we investigate the role of data diversity in enhancing the overall abilities of LLMs by empirically constructing contrastive data pools and theoretically deriving explanations. Building upon the insights gained, we propose a new method that gives the LLM a dual identity: an output model to cognitively probe and select data based on diversity reward, as well as an input model to be tuned with the selected data. Extensive experiments show that the proposed method notably boosts performance across domain-undetermined data and a series of foundational downstream tasks when applied to various advanced LLMs. We release our code and hope this study can shed light on the understanding of data diversity and advance feedback-driven data-model co-design for LLMs.

comment: Accepted by NeurIPS'25 main track. 47 pages, 21 figures, 32 tables

♻ ☆ Truncated Kernel Stochastic Gradient Descent with General Losses and Spherical Radial Basis Functions

Jinhui Bai, Andreas Christmann, Lei Shi

In this paper, we propose a novel kernel stochastic gradient descent (SGD) algorithm for large-scale supervised learning with general losses. Compared to traditional kernel SGD, our algorithm improves efficiency and scalability through an innovative regularization strategy. By leveraging the infinite series expansion of spherical radial basis functions, this strategy projects the stochastic gradient onto a finite-dimensional hypothesis space, which is adaptively scaled according to the bias-variance trade-off, thereby enhancing generalization performance. Based on a new estimation of the spectral structure of the kernel-induced covariance operator, we develop an analytical framework that unifies optimization and generalization analyses. We prove that both the last iterate and the suffix average converge at minimax-optimal rates, and we further establish optimal strong convergence in the reproducing kernel Hilbert space. Our framework accommodates a broad class of classical loss functions, including least-squares, Huber, and logistic losses. Moreover, the proposed algorithm significantly reduces computational complexity and achieves optimal storage complexity by incorporating coordinate-wise updates from linear SGD, thereby avoiding the costly pairwise operations typical of kernel SGD and enabling efficient processing of streaming data. Finally, extensive numerical experiments demonstrate the efficiency of our approach.

comment: 54 pages, 20 figures

♻ ☆ Experiments with Optimal Model Trees

Sabino Francesco Roselli, Eibe Frank

Model trees provide an appealing way to perform interpretable machine learning for both classification and regression problems. In contrast to ``classic'' decision trees with constant values in their leaves, model trees can use linear combinations of predictor variables in their leaf nodes to form predictions, which can help achieve higher accuracy and smaller trees. Typical algorithms for learning model trees from training data work in a greedy fashion, growing the tree in a top-down manner by recursively splitting the data into smaller and smaller subsets. Crucially, the selected splits are only locally optimal, potentially rendering the tree overly complex and less accurate than a tree whose structure is globally optimal for the training data. In this paper, we empirically investigate the effect of constructing globally optimal model trees for classification and regression with linear support vector machines at the leaf nodes. To this end, we present mixed-integer linear programming formulations to learn optimal trees, compute such trees for a large collection of benchmark data sets, and compare their performance against greedily grown model trees in terms of interpretability and accuracy. We also compare to classic optimal and greedily grown decision trees, random forests, and support vector machines. Our results show that optimal model trees can achieve competitive accuracy with very small trees. We also investigate the effect on the accuracy of replacing axis-parallel splits with multivariate ones, foregoing interpretability while potentially obtaining greater accuracy.

♻ ☆ Decoding for Punctured Convolutional and Turbo Codes: A Deep Learning Solution for Protocols Compliance

Yongli Yan, Linglong Dai

Neural network-based decoding methods show promise in enhancing error correction performance but face challenges with punctured codes. In particular, existing methods struggle to adapt to variable code rates or meet protocol compatibility requirements. This paper proposes a unified long short-term memory (LSTM)-based neural decoder for punctured convolutional and Turbo codes to address these challenges. The key component of the proposed LSTM-based neural decoder is puncturing-aware embedding, which integrates puncturing patterns directly into the neural network to enable seamless adaptation to different code rates. Moreover, a balanced bit error rate training strategy is designed to ensure the decoder's robustness across various code lengths, rates, and channels. In this way, the protocol compatibility requirement can be realized. Extensive simulations in both additive white Gaussian noise (AWGN) and Rayleigh fading channels demonstrate that the proposed neural decoder outperforms conventional decoding techniques, offering significant improvements in decoding accuracy and robustness.

♻ ☆ Beyond likelihood ratio bias: Nested multi-time-scale stochastic approximation for likelihood-free parameter estimation

Zehao Li, Zhouchen Lin, Yijie Peng

We study parameter inference in simulation-based stochastic models where the analytical form of the likelihood is unknown. The main difficulty is that score evaluation as a ratio of noisy Monte Carlo estimators induces bias and instability, which we overcome with a ratio-free nested multi-time-scale (NMTS) stochastic approximation (SA) method that simultaneously tracks the score and drives the parameter update. We provide a comprehensive theoretical analysis of the proposed NMTS algorithm for solving likelihood-free inference problems, including strong convergence, asymptotic normality, and convergence rates. We show that our algorithm can eliminate the original asymptotic bias $O\big(\sqrt{\frac{1}{N}}\big)$ and accelerate the convergence rate from $O\big(\beta_k+\sqrt{\frac{1}{N}}\big)$ to $O\big(\frac{\beta_k}{\alpha_k}+\sqrt{\frac{\alpha_k}{N}}\big)$, where $N$ is the fixed batch size, $\alpha_k$ and $\beta_k$ are decreasing step sizes with $\alpha_k$, $\beta_k$, $\beta_k/\alpha_k\rightarrow 0$. With proper choice of $\alpha_k$ and $\beta_k$, our convergence rates can match the optimal rate in the multi-time-scale SA literature. Numerical experiments demonstrate that our algorithm can improve the estimation accuracy by one to two orders of magnitude at the same computational cost, making it efficient for parameter estimation in stochastic systems.

♻ ☆ Chaos-based reinforcement learning with TD3

Toshitaka Matsuki, Yusuke Sakemi, Kazuyuki Aihara

Chaos-based reinforcement learning (CBRL) is a method in which the agent's internal chaotic dynamics drives exploration. However, the learning algorithms in CBRL have not been thoroughly developed in previous studies, nor have they incorporated recent advances in reinforcement learning. This study introduced Twin Delayed Deep Deterministic Policy Gradients (TD3), which is one of the state-of-the-art deep reinforcement learning algorithms that can treat deterministic and continuous action spaces, to CBRL. The validation results provide several insights. First, TD3 works as a learning algorithm for CBRL in a simple goal-reaching task. Second, CBRL agents with TD3 can autonomously suppress their exploratory behavior as learning progresses and resume exploration when the environment changes. Finally, examining the effect of the agent's chaoticity on learning shows that there exists a suitable range of chaos strength in the agent's model to flexibly switch between exploration and exploitation and adapt to environmental changes.

comment: Accepted for publication in Neural Networks

♻ ☆ Through the River: Understanding the Benefit of Schedule-Free Methods for Language Model Training NeurIPS 2025

Minhak Song, Beomhan Baek, Kwangjun Ahn, Chulhee Yun

As both model and dataset sizes continue to scale rapidly, conventional pretraining strategies with fixed compute budgets-such as cosine learning rate schedules-are increasingly inadequate for large-scale training. Recent alternatives, including warmup-stable-decay (WSD) schedules and weight averaging, offer greater flexibility. However, WSD relies on explicit decay phases to track progress, while weight averaging addresses this limitation at the cost of additional memory. In search of a more principled and scalable alternative, we revisit the Schedule-Free (SF) method [Defazio et al., 2024], which has shown strong empirical performance across diverse settings. We show that SF-AdamW effectively navigates the "river" structure of the loss landscape without decay phases or auxiliary averaging, making it particularly suitable for continuously scaling training workloads. To understand this behavior, we conduct a theoretical and empirical analysis of SF dynamics, revealing that it implicitly performs weight averaging without memory overhead. Guided by this analysis, we propose a refined variant of SF that improves robustness to momentum and performs better under large batch sizes, addressing key limitations of the original method. Together, these results establish SF as a practical, scalable, and theoretically grounded approach for language model training.

comment: Published at NeurIPS 2025

♻ ☆ VeriLLM: A Lightweight Framework for Publicly Verifiable Decentralized Inference

Ke Wang, Zishuo Zhao, Xinyuan Song, Bill Shi, Libin Xia, Chris Tong, Lynn Ai, Felix Qu, Eric Yang

Decentralized inference provides a scalable and resilient paradigm for serving large language models (LLMs), enabling distributed resource utilization and reducing reliance on centralized providers. However, in a permissionless environment without trusted nodes, ensuring the correctness of model outputs remains a core challenge. We introduce VeriLLM, a publicly verifiable protocol for decentralized LLM inference that achieves security under a one-honest-verifier assumption while maintaining practical efficiency. VeriLLM combines lightweight empirical rerunning with cryptographic commitments, allowing verifiers to validate results at approximately 1% of the underlying inference cost. To prevent verification bottlenecks, we design an isomorphic inference-verification architecture that multiplexes both inference and verification roles across the same GPU workers. This design (i) improves GPU utilization and overall throughput, (ii) enlarges the effective validator set, enhancing robustness and liveness, and (iii) enforces task indistinguishability to prevent node-specific optimizations or selective behavior. Through theoretical analysis and system-level evaluation, we show that VeriLLM achieves reliable public verifiability with minimal overhead, offering a practical foundation for trustworthy and scalable decentralized LLM inference.

comment: 20 pages, 4 figures, 6 tables

♻ ☆ A Convexity-dependent Two-Phase Training Algorithm for Deep Neural Networks

Tomas Hrycej, Bernhard Bermeitinger, Massimo Pavone, Götz-Henrik Wiegand, Siegfried Handschuh

The key task of machine learning is to minimize the loss function that measures the model fit to the training data. The numerical methods to do this efficiently depend on the properties of the loss function. The most decisive among these properties is the convexity or non-convexity of the loss function. The fact that the loss function can have, and frequently has, non-convex regions has led to a widespread commitment to non-convex methods such as Adam. However, a local minimum implies that, in some environment around it, the function is convex. In this environment, second-order minimizing methods such as the Conjugate Gradient (CG) give a guaranteed superlinear convergence. We propose a novel framework grounded in the hypothesis that loss functions in real-world tasks swap from initial non-convexity to convexity towards the optimum. This is a property we leverage to design an innovative two-phase optimization algorithm. The presented algorithm detects the swap point by observing the gradient norm dependence on the loss. In these regions, non-convex (Adam) and convex (CG) algorithms are used, respectively. Computing experiments confirm the hypothesis that this simple convexity structure is frequent enough to be practically exploited to substantially improve convergence and accuracy.

comment: Appeared on KDIR IC3K Conference 2025 (Best Paper Award). Published in "Proceedings of the 17th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management - Volume 1"

♻ ☆ MTL-KD: Multi-Task Learning Via Knowledge Distillation for Generalizable Neural Vehicle Routing Solver NeurIPS 2025

Yuepeng Zheng, Fu Luo, Zhenkun Wang, Yaoxin Wu, Yu Zhou

Multi-Task Learning (MTL) in Neural Combinatorial Optimization (NCO) is a promising approach to train a unified model capable of solving multiple Vehicle Routing Problem (VRP) variants. However, existing Reinforcement Learning (RL)-based multi-task methods can only train light decoder models on small-scale problems, exhibiting limited generalization ability when solving large-scale problems. To overcome this limitation, this work introduces a novel multi-task learning method driven by knowledge distillation (MTL-KD), which enables the efficient training of heavy decoder models with strong generalization ability. The proposed MTL-KD method transfers policy knowledge from multiple distinct RL-based single-task models to a single heavy decoder model, facilitating label-free training and effectively improving the model's generalization ability across diverse tasks. In addition, we introduce a flexible inference strategy termed Random Reordering Re-Construction (R3C), which is specifically adapted for diverse VRP tasks and further boosts the performance of the multi-task model. Experimental results on 6 seen and 10 unseen VRP variants with up to 1000 nodes indicate that our proposed method consistently achieves superior performance on both uniform and real-world benchmarks, demonstrating robust generalization abilities.

comment: Accepted at NeurIPS 2025

♻ ☆ Unified Error Correction Code Transformer with Low Complexity

Yongli Yan, Jieao Zhu, Tianyue Zheng, Zhuo Xu, Chao Jiang, Linglong Dai

Channel coding is vital for reliable sixth-generation (6G) data transmission, employing diverse error correction codes for various application scenarios. Traditional decoders require dedicated hardware for each code, leading to high hardware costs. Recently, artificial intelligence (AI)-driven approaches, such as the error correction code Transformer (ECCT) and its enhanced version, the foundation error correction code Transformer (FECCT), have been proposed to reduce the hardware cost by leveraging the Transformer to decode multiple codes. However, their excessively high computational complexity of $\mathcal{O}(N^2)$ due to the self-attention mechanism in the Transformer limits scalability, where $N$ represents the sequence length. To reduce computational complexity, we propose a unified Transformer-based decoder that handles multiple linear block codes within a single framework. Specifically, a standardized unit is employed to align code length and code rate across different code types, while a redesigned low-rank unified attention module, with computational complexity of $\mathcal{O}(N)$, is shared across various heads in the Transformer. Additionally, a sparse mask, derived from the parity-check matrix's sparsity, is introduced to enhance the decoder's ability to capture inherent constraints between information and parity-check bits, improving decoding accuracy and further reducing computational complexity by $86\%$. Extensive experimental results demonstrate that the proposed unified Transformer-based decoder outperforms existing methods and provides a high-performance, low-complexity solution for next-generation wireless communication systems.

♻ ☆ Optimal Online Change Detection via Random Fourier Features NeurIPS 2025

Florian Kalinke, Shakeel Gavioli-Akilagun

This article studies the problem of online non-parametric change point detection in multivariate data streams. We approach the problem through the lens of kernel-based two-sample testing and introduce a sequential testing procedure based on random Fourier features, running with logarithmic time complexity per observation and with overall logarithmic space complexity. The algorithm has two advantages compared to the state of the art. First, our approach is genuinely online, and no access to training data known to be from the pre-change distribution is necessary. Second, the algorithm does not require the user to specify a window parameter over which local tests are to be calculated. We prove strong theoretical guarantees on the algorithm's performance, including information-theoretic bounds demonstrating that the detection delay is optimal in the minimax sense. Numerical studies on real and synthetic data show that our algorithm is competitive with respect to the state of the art.

comment: Accepted for publication at NeurIPS 2025

♻ ☆ Rethinking Neural Combinatorial Optimization for Vehicle Routing Problems with Different Constraint Tightness Degrees NeurIPS 2025

Fu Luo, Yaoxin Wu, Zhi Zheng, Zhenkun Wang

Recent neural combinatorial optimization (NCO) methods have shown promising problem-solving ability without requiring domain-specific expertise. Most existing NCO methods use training and testing data with a fixed constraint value and lack research on the effect of constraint tightness on the performance of NCO methods. This paper takes the capacity-constrained vehicle routing problem (CVRP) as an example to empirically analyze the NCO performance under different tightness degrees of the capacity constraint. Our analysis reveals that existing NCO methods overfit the capacity constraint, and they can only perform satisfactorily on a small range of the constraint values but poorly on other values. To tackle this drawback of existing NCO methods, we develop an efficient training scheme that explicitly considers varying degrees of constraint tightness and proposes a multi-expert module to learn a generally adaptable solving strategy. Experimental results show that the proposed method can effectively overcome the overfitting issue, demonstrating superior performances on the CVRP and CVRP with time windows (CVRPTW) with various constraint tightness degrees.

comment: Accepted at NeurIPS 2025

♻ ☆ SPARKE: Scalable Prompt-Aware Diversity and Novelty Guidance in Diffusion Models via RKE Score

Mohammad Jalali, Haoyu Lei, Amin Gohari, Farzan Farnia

Diffusion models have demonstrated remarkable success in high-fidelity image synthesis and prompt-guided generative modeling. However, ensuring adequate diversity in generated samples of prompt-guided diffusion models remains a challenge, particularly when the prompts span a broad semantic spectrum and the diversity of generated data needs to be evaluated in a prompt-aware fashion across semantically similar prompts. Recent methods have introduced guidance via diversity measures to encourage more varied generations. In this work, we extend the diversity measure-based approaches by proposing the Scalable Prompt-Aware R\'eny Kernel Entropy Diversity Guidance (SPARKE) method for prompt-aware diversity guidance. SPARKE utilizes conditional entropy for diversity guidance, which dynamically conditions diversity measurement on similar prompts and enables prompt-aware diversity control. While the entropy-based guidance approach enhances prompt-aware diversity, its reliance on the matrix-based entropy scores poses computational challenges in large-scale generation settings. To address this, we focus on the special case of Conditional latent RKE Score Guidance, reducing entropy computation and gradient-based optimization complexity from the $O(n^3)$ of general entropy measures to $O(n)$. The reduced computational complexity allows for diversity-guided sampling over potentially thousands of generation rounds on different prompts. We numerically test the SPARKE method on several text-to-image diffusion models, demonstrating that the proposed method improves the prompt-aware diversity of the generated data without incurring significant computational costs. We release our code on the project page: https://mjalali.github.io/SPARKE

♻ ★ Efficient Regression-Based Training of Normalizing Flows for Boltzmann Generators ICML

Danyal Rehman, Oscar Davis, Jiarui Lu, Jian Tang, Michael Bronstein, Yoshua Bengio, Alexander Tong, Avishek Joey Bose

Simulation-free training frameworks have been at the forefront of the generative modelling revolution in continuous spaces, leading to large-scale diffusion and flow matching models. However, such modern generative models suffer from expensive inference, inhibiting their use in numerous scientific applications like Boltzmann Generators (BGs) for molecular conformations that require fast likelihood evaluation. In this paper, we revisit classical normalizing flows in the context of BGs that offer efficient sampling and likelihoods, but whose training via maximum likelihood is often unstable and computationally challenging. We propose Regression Training of Normalizing Flows (RegFlow), a novel and scalable regression-based training objective that bypasses the numerical instability and computational challenge of conventional maximum likelihood training in favour of a simple $\ell_2$-regression objective. Specifically, RegFlow maps prior samples under our flow to targets computed using optimal transport couplings or a pre-trained continuous normalizing flow (CNF). To enhance numerical stability, RegFlow employs effective regularization strategies such as a new forward-backward self-consistency loss that enjoys painless implementation. Empirically, we demonstrate that RegFlow unlocks a broader class of architectures that were previously intractable to train for BGs with maximum likelihood. We also show RegFlow exceeds the performance, computational cost, and stability of maximum likelihood training in equilibrium sampling in Cartesian coordinates of alanine dipeptide, tripeptide, and tetrapeptide, showcasing its potential in molecular systems.

comment: Preprint; ICML GenBio Best Paper Award 2025

♻ ☆ Learning to Insert for Constructive Neural Vehicle Routing Solver NeurIPS 2025

Fu Luo, Xi Lin, Mengyuan Zhong, Fei Liu, Zhenkun Wang, Jianyong Sun, Qingfu Zhang

Neural Combinatorial Optimisation (NCO) is a promising learning-based approach for solving Vehicle Routing Problems (VRPs) without extensive manual design. While existing constructive NCO methods typically follow an appending-based paradigm that sequentially adds unvisited nodes to partial solutions, this rigid approach often leads to suboptimal results. To overcome this limitation, we explore the idea of insertion-based paradigm and propose Learning to Construct with Insertion-based Paradigm (L2C-Insert), a novel learning-based method for constructive NCO. Unlike traditional approaches, L2C-Insert builds solutions by strategically inserting unvisited nodes at any valid position in the current partial solution, which can significantly enhance the flexibility and solution quality. The proposed framework introduces three key components: a novel model architecture for precise insertion position prediction, an efficient training scheme for model optimization, and an advanced inference technique that fully exploits the insertion paradigm's flexibility. Extensive experiments on both synthetic and real-world instances of the Travelling Salesman Problem (TSP) and Capacitated Vehicle Routing Problem (CVRP) demonstrate that L2C-Insert consistently achieves superior performance across various problem sizes.

comment: Accepted at NeurIPS 2025

♻ ★ Self-Evolving Curriculum for LLM Reasoning

Xiaoyin Chen, Jiarui Lu, Minsu Kim, Dinghuai Zhang, Jian Tang, Alexandre Piché, Nicolas Gontier, Yoshua Bengio, Ehsan Kamalloo

Reinforcement learning (RL) has proven effective for fine-tuning large language models (LLMs), significantly enhancing their reasoning abilities in domains such as mathematics and code generation. A crucial factor influencing RL fine-tuning success is the training curriculum: the order in which training problems are presented. While random curricula serve as common baselines, they remain suboptimal; manually designed curricula often rely heavily on heuristics, and online filtering methods can be computationally prohibitive. To address these limitations, we propose Self-Evolving Curriculum (SEC), an automatic curriculum learning method that learns a curriculum policy concurrently with the RL fine-tuning process. Our approach formulates curriculum selection as a non-stationary Multi-Armed Bandit problem, treating each problem category (e.g., difficulty level or problem type) as an individual arm. We leverage the absolute advantage from policy gradient methods as a proxy measure for immediate learning gain. At each training step, the curriculum policy selects categories to maximize this reward signal and is updated using the TD(0) method. Across three distinct reasoning domains: planning, inductive reasoning, and mathematics, our experiments demonstrate that SEC significantly improves models' reasoning capabilities, enabling better generalization to harder, out-of-distribution test problems. Additionally, our approach achieves better skill balance when fine-tuning simultaneously on multiple reasoning domains. These findings highlight SEC as a promising strategy for RL fine-tuning of LLMs.

♻ ☆ Improving Generalization of Neural Combinatorial Optimization for Vehicle Routing Problems via Test-Time Projection Learning

Yuanyao Chen, Rongsheng Chen, Fu Luo, Zhenkun Wang

Neural Combinatorial Optimization (NCO) has emerged as a promising learning-based paradigm for addressing Vehicle Routing Problems (VRPs) by minimizing the need for extensive manual engineering. While existing NCO methods, trained on small-scale instances (e.g., 100 nodes), have demonstrated considerable success on problems of similar scale, their performance significantly degrades when applied to large-scale scenarios. This degradation arises from the distributional shift between training and testing data, rendering policies learned on small instances ineffective for larger problems. To overcome this limitation, we introduce a novel learning framework driven by Large Language Models (LLMs). This framework learns a projection between the training and testing distributions, which is then deployed to enhance the scalability of the NCO model. Notably, unlike prevailing techniques that necessitate joint training with the neural network, our approach operates exclusively during the inference phase, obviating the need for model retraining. Extensive experiments demonstrate that our method enables a backbone model (trained on 100-node instances) to achieve superior performance on large-scale Traveling Salesman Problem (TSP) and Capacitated Vehicle Routing Problem (CVRP) of up to 100K nodes from diverse distributions.

comment: arXiv admin note: text overlap with arXiv:2505.24627

♻ ☆ Rough Path Signatures: Learning Neural RDEs for Portfolio Optimization

Ali Atiah Alzahrani

We tackle high-dimensional, path-dependent valuation and control and introduce a deep BSDE/2BSDE solver that couples truncated log-signatures with a neural rough differential equation (RDE) backbone. The architecture aligns stochastic analysis with sequence-to-path learning: a CVaR-tilted terminal objective targets left-tail risk, while an optional second-order (2BSDE) head supplies curvature estimates for risk-sensitive control. Under matched compute and parameter budgets, the method improves accuracy, tail fidelity, and training stability across Asian and barrier option pricing and portfolio control: at d=200 it achieves CVaR(0.99)=9.80% versus 12.00-13.10% for strong baselines, attains the lowest HJB residual (0.011), and yields the lowest RMSEs for Z and Gamma. Ablations over truncation depth, local windows, and tilt parameters confirm complementary gains from the sequence-to-path representation and the 2BSDE head. Taken together, the results highlight a bidirectional dialogue between stochastic analysis and modern deep learning: stochastic tools inform representations and objectives, while sequence-to-path models expand the class of solvable financial models at scale.

comment: Code available at: https://github.com/AliAtiah/SigRDE

♻ ☆ FESTA: Functionally Equivalent Sampling for Trust Assessment of Multimodal LLMs EMNLP

Debarpan Bhattacharya, Apoorva Kulkarni, Sriram Ganapathy

The accurate trust assessment of multimodal large language models (MLLMs) generated predictions, which can enable selective prediction and improve user confidence, is challenging due to the diverse multi-modal input paradigms. We propose Functionally Equivalent Sampling for Trust Assessment (FESTA), a multimodal input sampling technique for MLLMs, that generates an uncertainty measure based on the equivalent and complementary input samplings. The proposed task-preserving sampling approach for uncertainty quantification expands the input space to probe the consistency (through equivalent samples) and sensitivity (through complementary samples) of the model. FESTA uses only input-output access of the model (black-box), and does not require ground truth (unsupervised). The experiments are conducted with various off-the-shelf multi-modal LLMs, on both visual and audio reasoning tasks. The proposed FESTA uncertainty estimate achieves significant improvement (33.3% relative improvement for vision-LLMs and 29.6% relative improvement for audio-LLMs) in selective prediction performance, based on area-under-receiver-operating-characteristic curve (AUROC) metric in detecting mispredictions. The code implementation is open-sourced.

comment: Accepted in the Findings of EMNLP, 2025

♻ ☆ Shilling Recommender Systems by Generating Side-feature-aware Fake User Profiles

Yuanrong Wang, Yingpeng Du

Recommender systems (RS) greatly influence users' consumption decisions, making them attractive targets for malicious shilling attacks that inject fake user profiles to manipulate recommendations. Existing shilling methods can generate effective and stealthy fake profiles when training data only contain rating matrix, but they lack comprehensive solutions for scenarios where side features are present and utilized by the recommender. To address this gap, we extend the Leg-UP framework by enhancing the generator architecture to incorporate side features, enabling the generation of side-feature-aware fake user profiles. Experiments on benchmarks show that our method achieves strong attack performance while maintaining stealthiness.

♻ ☆ Seek in the Dark: Reasoning via Test-Time Instance-Level Policy Gradient in Latent Space

Hengli Li, Chenxi Li, Tong Wu, Xuekai Zhu, Yuxuan Wang, Zhaoxin Yu, Eric Hanchen Jiang, Song-Chun Zhu, Zixia Jia, Ying Nian Wu, Zilong Zheng

Reasoning ability, a core component of human intelligence, continues to pose a significant challenge for Large Language Models (LLMs) in the pursuit of AGI. Although model performance has improved under the training scaling law, significant challenges remain, particularly with respect to training algorithms, such as catastrophic forgetting, and the limited availability of novel training data. As an alternative, test-time scaling enhances reasoning performance by increasing test-time computation without parameter updating. Unlike prior methods in this paradigm focused on token space, we propose leveraging latent space for more effective reasoning and better adherence to the test-time scaling law. We introduce LatentSeek, a novel framework that enhances LLM reasoning through Test-Time Instance-level Adaptation (TTIA) within the model's latent space. Specifically, LatentSeek leverages policy gradient to iteratively update latent representations, guided by self-generated reward signals. LatentSeek is evaluated on a range of reasoning benchmarks, including GSM8K, MATH-500, and AIME2024, across multiple LLM architectures. Results show that LatentSeek consistently outperforms strong baselines, such as Chain-of-Thought prompting and fine-tuning-based methods. Furthermore, our analysis demonstrates that LatentSeek is highly efficient, typically converging within a few iterations for problems of average complexity, while also benefiting from additional iterations, thereby highlighting the potential of test-time scaling in the latent space. These findings position LatentSeek as a lightweight, scalable, and effective solution for enhancing the reasoning capabilities of LLMs.

♻ ☆ Improving the Euclidean Diffusion Generation of Manifold Data by Mitigating Score Function Singularity

Zichen Liu, Wei Zhang, Tiejun Li

Euclidean diffusion models have achieved remarkable success in generative modeling across diverse domains, and they have been extended to manifold cases in recent advances. Instead of explicitly utilizing the structure of special manifolds as studied in previous works, in this paper we investigate direct sampling of the Euclidean diffusion models for general manifold-structured data. We reveal the multiscale singularity of the score function in the ambient space, which hinders the accuracy of diffusion-generated samples. We then present an elaborate theoretical analysis of the singularity structure of the score function by decomposing it along the tangential and normal directions of the manifold. To mitigate the singularity and improve the sampling accuracy, we propose two novel methods: (1) Niso-DM, which reduces the scale discrepancies in the score function by utilizing a non-isotropic noise, and (2) Tango-DM, which trains only the tangential component of the score function using a tangential-only loss function. Numerical experiments demonstrate that our methods achieve superior performance on distributions over various manifolds with complex geometries.

♻ ☆ Send Less, Save More: Energy-Efficiency Benchmark of Embedded CNN Inference vs. Data Transmission in IoT

Benjamin Karic, Nina Herrmann, Jan Stenkamp, Paula Scharf, Fabian Gieseke, Angela Schwering

The integration of the Internet of Things (IoT) and Artificial Intelligence offers significant opportunities to enhance our ability to monitor and address ecological changes. As environmental challenges become increasingly pressing, the need for effective remote monitoring solutions is more critical than ever. A major challenge in designing IoT applications for environmental monitoring - particularly those involving image data - is to create energy-efficient IoT devices capable of long-term operation in remote areas with limited power availability. Advancements in the field of Tiny Machine Learning allow the use of Convolutional Neural Networks (CNNs) on resource-constrained, battery-operated microcontrollers. Since data transfer is energy-intensive, performing inference directly on microcontrollers to reduce the message size can extend the operational lifespan of IoT nodes. This work evaluates the use of common Low Power Wide Area Networks and compressed CNNs trained on domain specific datasets on an ESP32-S3. Our experiments demonstrate, among other things, that executing CNN inference on-device and transmitting only the results reduces the overall energy consumption by a factor of up to five compared to sending raw image data. These findings advocate the development of IoT applications with reduced carbon footprint and capable of operating autonomously in environmental monitoring scenarios by incorporating EmbeddedML.

comment: 11 Pages, Paper lists the categories for the ACM Computing Classification System

♻ ☆ AANet: Virtual Screening under Structural Uncertainty via Alignment and Aggregation NeurIPS 2025

Wenyu Zhu, Jianhui Wang, Bowen Gao, Yinjun Jia, Haichuan Tan, Ya-Qin Zhang, Wei-Ying Ma, Yanyan Lan

Virtual screening (VS) is a critical component of modern drug discovery, yet most existing methods--whether physics-based or deep learning-based--are developed around holo protein structures with known ligand-bound pockets. Consequently, their performance degrades significantly on apo or predicted structures such as those from AlphaFold2, which are more representative of real-world early-stage drug discovery, where pocket information is often missing. In this paper, we introduce an alignment-and-aggregation framework to enable accurate virtual screening under structural uncertainty. Our method comprises two core components: (1) a tri-modal contrastive learning module that aligns representations of the ligand, the holo pocket, and cavities detected from structures, thereby enhancing robustness to pocket localization error; and (2) a cross-attention based adapter for dynamically aggregating candidate binding sites, enabling the model to learn from activity data even without precise pocket annotations. We evaluated our method on a newly curated benchmark of apo structures, where it significantly outperforms state-of-the-art methods in blind apo setting, improving the early enrichment factor (EF1%) from 11.75 to 37.19. Notably, it also maintains strong performance on holo structures. These results demonstrate the promise of our approach in advancing first-in-class drug discovery, particularly in scenarios lacking experimentally resolved protein-ligand complexes. Our implementation is publicly available at https://github.com/Wiley-Z/AANet.

comment: Accepted at NeurIPS 2025

♻ ☆ Parallel Unlearning in Inherited Model Networks

Xiao Liu, Mingyuan Li, Guangsheng Yu, Lixiang Li, Haipeng Peng, Ren Ping Liu

Unlearning is challenging in generic learning frameworks with the continuous growth and updates of models exhibiting complex inheritance relationships. This paper presents a novel unlearning framework that enables fully parallel unlearning among models exhibiting inheritance. We use a chronologically Directed Acyclic Graph (DAG) to capture various unlearning scenarios occurring in model inheritance networks. Central to our framework is the Fisher Inheritance Unlearning (FIUn) method, designed to enable efficient parallel unlearning within the DAG. FIUn utilizes the Fisher Information Matrix (FIM) to assess the significance of model parameters for unlearning tasks and adjusts them accordingly. To handle multiple unlearning requests simultaneously, we propose the Merging-FIM (MFIM) function, which consolidates FIMs from multiple upstream models into a unified matrix. This design supports all unlearning scenarios captured by the DAG, enabling one-shot removal of inherited knowledge while significantly reducing computational overhead. Experiments confirm the effectiveness of our unlearning framework. For single-class tasks, it achieves complete unlearning with 0% accuracy for unlearned labels while maintaining 94.53% accuracy for retained labels. For multi-class tasks, the accuracy is 1.07% for unlearned labels and 84.77% for retained labels. Our framework accelerates unlearning by 99% compared to alternative methods. Code is in https://github.com/MJLee00/Parallel-Unlearning-in-Inherited-Model-Networks.

♻ ☆ Similarity-Distance-Magnitude Activations

Allen Schmaltz

We introduce the Similarity-Distance-Magnitude (SDM) activation function, a more robust and interpretable formulation of the standard softmax activation function, adding Similarity (i.e., correctly predicted depth-matches into training) awareness and Distance-to-training-distribution awareness to the existing output Magnitude (i.e., decision-boundary) awareness, and enabling interpretability-by-exemplar via dense matching. We further introduce the SDM estimator, based on a data-driven partitioning of the class-wise empirical CDFs via the SDM activation, to control the class- and prediction-conditional accuracy among selective classifications. When used as the final-layer activation over pre-trained language models for selective classification, the SDM estimator is more robust to co-variate shifts and out-of-distribution inputs than existing calibration methods using softmax activations, while remaining informative over in-distribution data.

comment: 18 pages, 5 tables, 1 algorithm. arXiv admin note: substantial text overlap with arXiv:2502.20167

♻ ☆ A Robust and Non-Iterative Tensor Decomposition Method with Automatic Thresholding

Hiroki Hasegawa, Yukihiko Okada

Recent advances in IoT and biometric sensing technologies have led to the generation of massive and high-dimensional tensor data, yet achieving accurate and efficient low-rank approximation remains a major challenge. Existing tensor decomposition methods typically require prior specification of the tensor rank and rely on iterative optimization, which often results in heavy computational costs and dependence on the analyst's expertise. In this study, we propose a novel low-rank approximation method for tensor data that requires neither prior rank specification nor iterative optimization. The proposed method performs statistical singular value hard thresholding on the mode-wise unfolded matrices to automatically extract only statistically significant components, thereby achieving noise reduction while preserving the intrinsic tensor structure. Theoretically, the optimal threshold for each mode is derived based on the asymptotic properties of the Mar\v{c}enko--Pastur distribution. Simulation experiments demonstrate that the proposed method outperforms conventional approaches such as Higher-Order Singular Value Decomposition, Higher-Order Orthogonal Iteration, and Tucker-L2E in terms of both estimation accuracy and computational efficiency. These results indicate that our method provides an effective and theoretically grounded framework for automatic, non-iterative, and analyst-independent tensor decomposition.

comment: 25 pages, 3 figures

♻ ☆ Buffer layers for Test-Time Adaptation NeurIPS 2025

Hyeongyu Kim, Geonhui Han, Dosik Hwang

In recent advancements in Test Time Adaptation (TTA), most existing methodologies focus on updating normalization layers to adapt to the test domain. However, the reliance on normalization-based adaptation presents key challenges. First, normalization layers such as Batch Normalization (BN) are highly sensitive to small batch sizes, leading to unstable and inaccurate statistics. Moreover, normalization-based adaptation is inherently constrained by the structure of the pre-trained model, as it relies on training-time statistics that may not generalize well to unseen domains. These issues limit the effectiveness of normalization-based TTA approaches, especially under significant domain shift. In this paper, we introduce a novel paradigm based on the concept of a Buffer layer, which addresses the fundamental limitations of normalization layer updates. Unlike existing methods that modify the core parameters of the model, our approach preserves the integrity of the pre-trained backbone, inherently mitigating the risk of catastrophic forgetting during online adaptation. Through comprehensive experimentation, we demonstrate that our approach not only outperforms traditional methods in mitigating domain shift and enhancing model robustness, but also exhibits strong resilience to forgetting. Furthermore, our Buffer layer is modular and can be seamlessly integrated into nearly all existing TTA frameworks, resulting in consistent performance improvements across various architectures. These findings validate the effectiveness and versatility of the proposed solution in real-world domain adaptation scenarios. The code is available at https://github.com/hyeongyu-kim/Buffer_TTA.

comment: Accepted at NeurIPS 2025

♻ ☆ Generalized Linear Bandits: Almost Optimal Regret with One-Pass Update NeurIPS 2025

Yu-Jie Zhang, Sheng-An Xu, Peng Zhao, Masashi Sugiyama

We study the generalized linear bandit (GLB) problem, a contextual multi-armed bandit framework that extends the classical linear model by incorporating a non-linear link function, thereby modeling a broad class of reward distributions such as Bernoulli and Poisson. While GLBs are widely applicable to real-world scenarios, their non-linear nature introduces significant challenges in achieving both computational and statistical efficiency. Existing methods typically trade off between two objectives, either incurring high per-round costs for optimal regret guarantees or compromising statistical efficiency to enable constant-time updates. In this paper, we propose a jointly efficient algorithm that attains a nearly optimal regret bound with $\mathcal{O}(1)$ time and space complexities per round. The core of our method is a tight confidence set for the online mirror descent (OMD) estimator, which is derived through a novel analysis that leverages the notion of mix loss from online prediction. The analysis shows that our OMD estimator, even with its one-pass updates, achieves statistical efficiency comparable to maximum likelihood estimation, thereby leading to a jointly efficient optimistic method.

comment: NeurIPS 2025

♻ ☆ Holographic Transformers for Complex-Valued Signal Processing: Integrating Phase Interference into Self-Attention

Enhao Huang, Zhiyu Zhang, Tianxiang Xu, Chunshu Xia, Kaichun Hu, Yuchen Yang, Tongtong Pan, Dong Dong, Zhan Qin

Complex-valued signals encode both amplitude and phase, yet most deep models treat attention as real-valued correlation, overlooking interference effects. We introduce the Holographic Transformer, a physics-inspired architecture that incorporates wave interference principles into self-attention. Holographic attention modulates interactions by relative phase and coherently superimposes values, ensuring consistency between amplitude and phase. A dual-headed decoder simultaneously reconstructs the input and predicts task outputs, preventing phase collapse when losses prioritize magnitude over phase. We demonstrate that holographic attention implements a discrete interference operator and maintains phase consistency under linear mixing. Experiments on PolSAR image classification and wireless channel prediction show strong performance, achieving high classification accuracy and F1 scores, low regression error, and increased robustness to phase perturbations. These results highlight that enforcing physical consistency in attention leads to generalizable improvements in complex-valued learning and provides a unified, physics-based framework for coherent signal modeling. The code is available at https://github.com/EonHao/Holographic-Transformers.

♻ ☆ Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems

Christian Walder, Deep Karkhanis

Reinforcement Learning (RL) algorithms sample multiple n>1 solution attempts for each problem and reward them independently. This optimizes for pass@1 performance and prioritizes the strength of isolated samples at the expense of the diversity and collective utility of sets of samples. This under-utilizes the sampling capacity, limiting exploration and eventual improvement on harder examples. As a fix, we propose Pass-at-k Policy Optimization (PKPO), a transformation on the final rewards which leads to direct optimization of pass@k performance, thus optimizing for sets of samples that maximize reward when considered jointly. Our contribution is to derive novel low variance unbiased estimators for pass@k and its gradient, in both the binary and continuous reward settings. We show optimization with our estimators reduces to standard RL with rewards that have been jointly transformed by a stable and efficient transformation function. While previous efforts are restricted to k=n, ours is the first to enable robust optimization of pass@k for any arbitrary k <= n. Moreover, instead of trading off pass@1 performance for pass@k gains, our method allows annealing k during training, optimizing both metrics and often achieving strong pass@1 numbers alongside significant pass@k gains. We validate our reward transformations on toy experiments, which reveal the variance reducing properties of our formulations. We also include real-world examples using the open-source LLM, GEMMA-2. We find that our transformation effectively optimizes for the target k. Furthermore, higher k values enable solving more and harder problems, while annealing k boosts both the pass@1 and pass@k . Crucially, for challenging task sets where conventional pass@1 optimization stalls, our pass@k approach unblocks learning, likely due to better exploration by prioritizing joint utility over the utility of individual samples.

♻ ☆ Accurate predictive model of band gap with selected important features based on explainable machine learning

Joohwi Lee, Kaito Miyamoto

In the rapidly advancing field of materials informatics, nonlinear machine learning models have demonstrated exceptional predictive capabilities for material properties. However, their black-box nature limits interpretability, and they may incorporate features that do not contribute to, or even deteriorate, model performance. This study employs explainable ML (XML) techniques, including permutation feature importance and the SHapley Additive exPlanation, applied to a pristine support vector regression model designed to predict band gaps at the GW level using 18 input features. Guided by XML-derived individual feature importance, a simple framework is proposed to construct reduced-feature predictive models. Model evaluations indicate that an XML-guided compact model, consisting of the top five features, achieves comparable accuracy to the pristine model on in-domain datasets (0.254 vs. 0.247 eV) while demonstrating superior generalization with lower prediction errors on out-of-domain data (0.461 vs. 0.341 eV). Additionally, the study underscores the necessity for eliminating strongly correlated features (correlation coefficient greater than 0.8) to prevent misinterpretation and overestimation of feature importance before applying XML. This study highlights XML's effectiveness in developing simplified yet highly accurate machine learning models by clarifying feature roles, thereby reducing computational costs for feature acquisition and enhancing model trustworthiness for materials discovery.

comment: 9 pages, 3 figures, SI is included

♻ ☆ Neural Networks for Learnable and Scalable Influence Estimation of Instruction Fine-Tuning Data

Ishika Agarwal, Dilek Hakkani-Tür

Influence functions provide crucial insights into model training, but existing methods suffer from large computational costs and limited generalization. Particularly, recent works have proposed various metrics and algorithms to calculate the influence of data using language models, which do not scale well with large models and datasets. This is because of the expensive forward and backward passes required for computation, substantial memory requirements to store large models, and poor generalization of influence estimates to new data. In this paper, we explore the use of small neural networks -- which we refer to as the InfluenceNetwork -- to estimate influence values, achieving up to 99% cost reduction. Our evaluation demonstrates that influence values can be estimated with models just 0.0027% the size of full language models (we use 7B and 8B versions). We apply our algorithm of estimating influence values (called NN-CIFT: Neural Networks for effiCient Instruction Fine-Tuning) to the downstream task of subset selection for general instruction fine-tuning. In our study, we include four state-of-the-art influence functions and show no compromise in performance, despite large speedups, between NN-CIFT and the original influence functions. We provide an in-depth hyperparameter analyses of NN-CIFT. The code for our method can be found here: https://github.com/agarwalishika/NN-CIFT.

♻ ☆ Localized Kernel Projection Outlyingness: A Two-Stage Approach for Multi-Modal Outlier Detection

Akira Tamamori

This paper presents Two-Stage LKPLO, a novel multi-stage outlier detection framework that overcomes the coexisting limitations of conventional projection-based methods: their reliance on a fixed statistical metric and their assumption of a single data structure. Our framework uniquely synthesizes three key concepts: (1) a generalized loss-based outlyingness measure (PLO) that replaces the fixed metric with flexible, adaptive loss functions like our proposed SVM-like loss; (2) a global kernel PCA stage to linearize non-linear data structures; and (3) a subsequent local clustering stage to handle multi-modal distributions. Comprehensive 5-fold cross-validation experiments on 10 benchmark datasets, with automated hyperparameter optimization, demonstrate that Two-Stage LKPLO achieves state-of-the-art performance. It significantly outperforms strong baselines on datasets with challenging structures where existing methods fail, most notably on multi-cluster data (Optdigits) and complex, high-dimensional data (Arrhythmia). Furthermore, an ablation study empirically confirms that the synergistic combination of both the kernelization and localization stages is indispensable for its superior performance. This work contributes a powerful new tool for a significant class of outlier detection problems and underscores the importance of hybrid, multi-stage architectures.

comment: 10 pages, 4 figures; submitted to The IEICE Transactions on Information and Systems

♻ ☆ MMEdge: Accelerating On-device Multimodal Inference via Pipelined Sensing and Encoding

Runxi Huang, Mingxuan Yu, Mingyu Tsoi, Xiaomin Ouyang

Real-time multimodal inference on resource-constrained edge devices is essential for applications such as autonomous driving, human-computer interaction, and mobile health. However, prior work often overlooks the tight coupling between sensing dynamics and model execution, as well as the complex inter-modality dependencies. In this paper, we propose MMEdge, an new on-device multi-modal inference framework based on pipelined sensing and encoding. Instead of waiting for complete sensor inputs, MMEdge decomposes the entire inference process into a sequence of fine-grained sensing and encoding units, allowing computation to proceed incrementally as data arrive. MMEdge also introduces a lightweight but effective temporal aggregation module that captures rich temporal dynamics across different pipelined units to maintain accuracy performance. Such pipelined design also opens up opportunities for fine-grained cross-modal optimization and early decision-making during inference. To further enhance system performance under resource variability and input data complexity, MMEdge incorporates an adaptive multimodal configuration optimizer that dynamically selects optimal sensing and model configurations for each modality under latency constraints, and a cross-modal speculative skipping mechanism that bypasses future units of slower modalities when early predictions reach sufficient confidence. We evaluate MMEdge using two public multimodal datasets and deploy it on a real-world unmanned aerial vehicle (UAV)-based multimodal testbed. The results show that MMEdge significantly reduces end-to-end latency while maintaining high task accuracy across various system and data dynamics.

comment: Code available at: https://github.com/HKUST-MINSys-Lab/MMEdge. Accepted by SenSys 2026

☆ MORE: Multi-Organ Medical Image REconstruction Dataset

Shaokai Wu, Yapan Guo, Yanbiao Ji, Jing Tong, Yuxiang Lu, Mei Li, Suizhi Huang, Yue Ding, Hongtao Lu

CT reconstruction provides radiologists with images for diagnosis and treatment, yet current deep learning methods are typically limited to specific anatomies and datasets, hindering generalization ability to unseen anatomies and lesions. To address this, we introduce the Multi-Organ medical image REconstruction (MORE) dataset, comprising CT scans across 9 diverse anatomies with 15 lesion types. This dataset serves two key purposes: (1) enabling robust training of deep learning models on extensive, heterogeneous data, and (2) facilitating rigorous evaluation of model generalization for CT reconstruction. We further establish a strong baseline solution that outperforms prior approaches under these challenging conditions. Our results demonstrate that: (1) a comprehensive dataset helps improve the generalization capability of models, and (2) optimization-based methods offer enhanced robustness for unseen anatomies. The MORE dataset is freely accessible under CC-BY-NC 4.0 at our project page https://more-med.github.io/

comment: Accepted to ACMMM 2025

☆ Unveiling Intrinsic Text Bias in Multimodal Large Language Models through Attention Key-Space Analysis

Xinhan Zheng, Huyu Wu, Xueting Wang, Haiyun Jiang

Multimodal large language models (MLLMs) exhibit a pronounced preference for textual inputs when processing vision-language data, limiting their ability to reason effectively from visual evidence. Unlike prior studies that attribute this text bias to external factors such as data imbalance or instruction tuning, we propose that the bias originates from the model's internal architecture. Specifically, we hypothesize that visual key vectors (Visual Keys) are out-of-distribution (OOD) relative to the text key space learned during language-only pretraining. Consequently, these visual keys receive systematically lower similarity scores during attention computation, leading to their under-utilization in the context representation. To validate this hypothesis, we extract key vectors from LLaVA and Qwen2.5-VL and analyze their distributional structures using qualitative (t-SNE) and quantitative (Jensen-Shannon divergence) methods. The results provide direct evidence that visual and textual keys occupy markedly distinct subspaces within the attention space. The inter-modal divergence is statistically significant, exceeding intra-modal variation by several orders of magnitude. These findings reveal that text bias arises from an intrinsic misalignment within the attention key space rather than solely from external data factors.

☆ AdSum: Two-stream Audio-visual Summarization for Automated Video Advertisement Clipping

Wen Xie, Yanjun Zhu, Gijs Overgoor, Yakov Bart, Agata Lapedriza Garcia, Sarah Ostadabbas

Advertisers commonly need multiple versions of the same advertisement (ad) at varying durations for a single campaign. The traditional approach involves manually selecting and re-editing shots from longer video ads to create shorter versions, which is labor-intensive and time-consuming. In this paper, we introduce a framework for automated video ad clipping using video summarization techniques. We are the first to frame video clipping as a shot selection problem, tailored specifically for advertising. Unlike existing general video summarization methods that primarily focus on visual content, our approach emphasizes the critical role of audio in advertising. To achieve this, we develop a two-stream audio-visual fusion model that predicts the importance of video frames, where importance is defined as the likelihood of a frame being selected in the firm-produced short ad. To address the lack of ad-specific datasets, we present AdSum204, a novel dataset comprising 102 pairs of 30-second and 15-second ads from real advertising campaigns. Extensive experiments demonstrate that our model outperforms state-of-the-art methods across various metrics, including Average Precision, Area Under Curve, Spearman, and Kendall.

comment: Accepted at 32nd International Conference on MultiMedia Modeling

☆ Contribution-Guided Asymmetric Learning for Robust Multimodal Fusion under Imbalance and Noise

Zijing Xu, Yunfeng Kou, Kunming Wu, Hong Liu

Multimodal learning faces two major challenges: modality imbalance and data noise, which significantly affect the robustness and generalization ability of models. Existing methods achieve modality balance by suppressing dominant modalities, but they neglect the inherent differences in the information value between modalities, potentially leading to convergence to suboptimal solutions. This paper proposes an innovative modality compression paradigm, Contribution-Guided Asymmetric Learning (CAL), which aims to enhance the contribution of high-contribution modalities while compressing weak modalities to increase their contribution, allowing both to improve the performance of multimodal information fusion. CAL is based on a modality contribution metric W^m combining the information quantity I(m) and confidence D(m), and it designs an asymmetric gradient acceleration mechanism and a contribution-aware Asymmetric Information Bottleneck (AIB) compression mechanism. The former accelerates the gradient update of modalities, while the latter dynamically compresses the noise of low-contribution modalities. On five benchmark datasets, including emotion recognition, scene recognition, and event localization tasks, CAL has shown outstanding performance in imbalanced fusion tasks and noise robustness tests. On CREMA-D, KS, and AVE, CAL achieves 79.30%, 74.82%, and 74.21% accuracy, significantly outperforming the existing state-of-the-art model ARL. In high-noise robustness tests, CAL also achieved leading performance under various attack strategies on the MVSA-Single and NYUD2 datasets. These results validate the significant advantages of CAL in modality imbalance and noise interference. CAL, as a flexible and efficient framework, is easy to transfer to other tasks and has broad adaptability and potential application prospects.

♻ ☆ Quality Over Quantity? LLM-Based Curation for a Data-Efficient Audio-Video Foundation Model

Ali Vosoughi, Dimitra Emmanouilidou, Hannes Gamper

Integrating audio and visual data for training multimodal foundational models remains a challenge. The Audio-Video Vector Alignment (AVVA) framework addresses this by considering AV scene alignment beyond mere temporal synchronization, and leveraging Large Language Models (LLMs) for data curation. AVVA implements a scoring mechanism for selecting aligned training data segments. It integrates Whisper, a speech-based foundation model, for audio and DINOv2 for video analysis in a dual-encoder structure with contrastive learning on AV pairs. Evaluations on AudioCaps, VALOR, and VGGSound demonstrate the effectiveness of the proposed model architecture and data curation approach. AVVA achieves a significant improvement in top-k accuracies for video-to-audio retrieval on all datasets compared to DenseAV, while using only 192 hrs of curated training data. Furthermore, an ablation study indicates that the data curation process effectively trades data quality for data quantity, yielding increases in top-k retrieval accuracies on AudioCaps, VALOR, and VGGSound, compared to training on the full spectrum of uncurated data.

comment: 5 pages, 5 figures, 2 tables. Accepted at EUSIPCO 2025

♻ ☆ ARECHO: Autoregressive Evaluation via Chain-Based Hypothesis Optimization for Speech Multi-Metric Estimation NeurIPS 2025

Jiatong Shi, Yifan Cheng, Bo-Hao Su, Hye-jin Shim, Jinchuan Tian, Samuele Cornell, Yiwen Zhao, Siddhant Arora, Shinji Watanabe

Speech signal analysis poses significant challenges, particularly in tasks such as speech quality evaluation and profiling, where the goal is to predict multiple perceptual and objective metrics. For instance, metrics like PESQ (Perceptual Evaluation of Speech Quality), STOI (Short-Time Objective Intelligibility), and MOS (Mean Opinion Score) each capture different aspects of speech quality. However, these metrics often have different scales, assumptions, and dependencies, making joint estimation non-trivial. To address these issues, we introduce ARECHO (Autoregressive Evaluation via Chain-based Hypothesis Optimization), a chain-based, versatile evaluation system for speech assessment grounded in autoregressive dependency modeling. ARECHO is distinguished by three key innovations: (1) a comprehensive speech information tokenization pipeline; (2) a dynamic classifier chain that explicitly captures inter-metric dependencies; and (3) a two-step confidence-oriented decoding algorithm that enhances inference reliability. Experiments demonstrate that ARECHO significantly outperforms the baseline framework across diverse evaluation scenarios, including enhanced speech analysis, speech generation evaluation, and, noisy speech evaluation. Furthermore, its dynamic dependency modeling improves interpretability by capturing inter-metric relationships. Across tasks, ARECHO offers reference-free evaluation using its dynamic classifier chain to support subset queries (single or multiple metrics) and reduces error propagation via confidence-oriented decoding.

comment: NeurIPS 2025 Spotlight

♻ ☆ Dependency Structure Augmented Contextual Scoping Framework for Multimodal Aspect-Based Sentiment Analysis

Hao Liu, Lijun He, Jiaxi Liang, Zhihan Ren, Haixia Bi, Fan Li

Multimodal Aspect-Based Sentiment Analysis (MABSA) seeks to extract fine-grained information from image-text pairs to identify aspect terms and determine their sentiment polarity. However, existing approaches often fall short in simultaneously addressing three core challenges: Sentiment Cue Perception (SCP), Multimodal Information Misalignment (MIM), and Semantic Noise Elimination (SNE). To overcome these limitations, we propose DASCO (\textbf{D}ependency Structure \textbf{A}ugmented \textbf{Sco}ping Framework), a fine-grained scope-oriented framework that enhances aspect-level sentiment reasoning by leveraging dependency parsing trees. First, we designed a multi-task pretraining strategy for MABSA on our base model, combining aspect-oriented enhancement, image-text matching, and aspect-level sentiment-sensitive cognition. This improved the model's perception of aspect terms and sentiment cues while achieving effective image-text alignment, addressing key challenges like SCP and MIM. Furthermore, we incorporate dependency trees as syntactic branch combining with semantic branch, guiding the model to selectively attend to critical contextual elements within a target-specific scope while effectively filtering out irrelevant noise for addressing SNE problem. Extensive experiments on two benchmark datasets across three subtasks demonstrate that DASCO achieves state-of-the-art performance in MABSA, with notable gains in JMASA (+2.3\% F1 and +3.5\% precision on Twitter2015). The source code is available at https://github.com/LHaoooo/DASCO .

♻ ☆ ReCon-GS: Continuum-Preserved Gaussian Streaming for Fast and Compact Reconstruction of Dynamic Scenes NeurIPS 2025

Jiaye Fu, Qiankun Gao, Chengxiang Wen, Yanmin Wu, Siwei Ma, Jiaqi Zhang, Jian Zhang

Online free-viewpoint video (FVV) reconstruction is challenged by slow per-frame optimization, inconsistent motion estimation, and unsustainable storage demands. To address these challenges, we propose the Reconfigurable Continuum Gaussian Stream, dubbed ReCon-GS, a novel storage-aware framework that enables high fidelity online dynamic scene reconstruction and real-time rendering. Specifically, we dynamically allocate multi-level Anchor Gaussians in a density-adaptive fashion to capture inter-frame geometric deformations, thereby decomposing scene motion into compact coarse-to-fine representations. Then, we design a dynamic hierarchy reconfiguration strategy that preserves localized motion expressiveness through on-demand anchor re-hierarchization, while ensuring temporal consistency through intra-hierarchical deformation inheritance that confines transformation priors to their respective hierarchy levels. Furthermore, we introduce a storage-aware optimization mechanism that flexibly adjusts the density of Anchor Gaussians at different hierarchy levels, enabling a controllable trade-off between reconstruction fidelity and memory usage. Extensive experiments on three widely used datasets demonstrate that, compared to state-of-the-art methods, ReCon-GS improves training efficiency by approximately 15% and achieves superior FVV synthesis quality with enhanced robustness and stability. Moreover, at equivalent rendering quality, ReCon-GS slashes memory requirements by over 50% compared to leading state-of-the-art methods.

comment: Published in NeurIPS 2025

♻ ☆ TRUST-VL: An Explainable News Assistant for General Multimodal Misinformation Detection EMNLP 2025

Zehong Yan, Peng Qi, Wynne Hsu, Mong Li Lee

Multimodal misinformation, encompassing textual, visual, and cross-modal distortions, poses an increasing societal threat that is amplified by generative AI. Existing methods typically focus on a single type of distortion and struggle to generalize to unseen scenarios. In this work, we observe that different distortion types share common reasoning capabilities while also requiring task-specific skills. We hypothesize that joint training across distortion types facilitates knowledge sharing and enhances the model's ability to generalize. To this end, we introduce TRUST-VL, a unified and explainable vision-language model for general multimodal misinformation detection. TRUST-VL incorporates a novel Question-Aware Visual Amplifier module, designed to extract task-specific visual features. To support training, we also construct TRUST-Instruct, a large-scale instruction dataset containing 198K samples featuring structured reasoning chains aligned with human fact-checking workflows. Extensive experiments on both in-domain and zero-shot benchmarks demonstrate that TRUST-VL achieves state-of-the-art performance, while also offering strong generalization and interpretability.

comment: EMNLP 2025 Oral; Project Homepage: https://yanzehong.github.io/trust-vl/

♻ ☆ PureKV: Plug-and-Play KV Cache Optimization with Spatial-Temporal Sparse Attention for Vision-Language Large Models

Zhonghua Jiang, Kunxi Li, Yiyun Zhou, Sihao Liu, Zhaode Wang, Chengfei lv, Shengyu Zhang

Vision-Language Large Models (VLLMs) face significant efficiency challenges when processing high-resolution inputs. The quadratic complexity in attention and autoregressive generation, as well as the constantly growing key value (KV) cache size, severely hinder the prefilling and decoding stages. Recent efforts have attempted to compress KV cache by identifying and pruning KV cache of less important tokens, but these methods typically rely on attention scores to estimate token importance, making them incompatible with efficient attention mechanisms such as FlashAttention and Sparse Attention, which do not explicitly compute attention matrices. Moreover, existing methods overlook how sparse attention, while accelerating the prefilling stage, alters the information structure of the KV cache, thereby compromising the effectiveness of downstream KV cache compression strategies. To address this issue, we propose PureKV, a plug-and-play framework for joint optimization of sparse attention and KV cache compression. We first introduce a KV cache compression strategy that is fully compatible with efficient attention accelerators. Our method utilizes lower layer attention scores to estimate the importance of high layers' KV cache, enabling active pruning without compromising accuracy. In addition, we have designed a Spatial-Temporal Sparse Attention (ST-SpAttn) module specifically tailored for video KV cache compression algorithms. This module combines spatial and temporal attention sparsity to improve the compression efficiency of KV cache optimization algorithms by purifying spatial noise and temporal redundancy in KV cache. At the same time, ST-SpAttn also accelerated the prefilling stage of VLLMs. Extensive experiments on VLLMs (VideoLLaMA2, Qwen2.5-VL) have shown that PureKV achieves 5.0 times KV cache compression and 3.16 times prefill acceleration, with negligible quality degradation.